WO2006128140A2

WO2006128140A2 - Gene methylation and expression

Info

Publication number: WO2006128140A2
Application number: PCT/US2006/020843
Authority: WO
Inventors: Kornelia Polyak; Min Hu; Noga Qimron; Jun Yao
Original assignee: Dana-Farber Cancer Institute, Inc.
Priority date: 2005-05-27
Filing date: 2006-05-30
Publication date: 2006-11-30
Also published as: JP2008545417A; EP1885919A4; US20090280478A1; EP1885919B1; EP1885919A2; CA2609512A1; WO2006128140A3; US9556430B2; ES2406943T3; JP5352232B2

Abstract

The invention provides a method of analyzing the methylation status of all or part of an entire genome. Moreover, the invention features methods of and reagents for characterizing biological cells containing DNA that is susceptible to methylation. Such methods include methods of diagnosing cancer, e.g., breast cancer.

Description

Gene Methylation and Expression

CROSS REFERENCE TO RELATED APPLICATIONS

This application claims priority to U.S. Provisional Application No. 60/685,104, filed May 27, 2005. The entire content of the prior application is incorporated herein by reference in its entirety.

STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH

The research described in this application was supported in part by grants (Nos. CA89393 and CA94074) from the National Cancer Institute of the National Institutes of Health, and grants (Nos. DAMD 17-02-1-0692 and W8IXWH-04- 1-0452) from the Department of Defense. Thus the government has certain rights in the invention.

TECHNICAL FIELD

This invention relates to epigenetic gene regulation, and more particularly to DNA methylation and its effect on gene expression, and its use as a marker of a particular cell type and/or disease state.

BACKGROUND

Epigenetic changes (e.g., changes in the levels of DNA methylation), as well as genetic changes, can be detected in cancer cells and stromal cells within tumors. In order to develop more discriminatory diagnostic methods and more effective therapeutic methods it is important that these epigenetic effects be defined and characterized.

SUMMARY

The inventors have developed a method of assessing the level of methylation in an entire, or part of a, genome. They call this method Methylation Specific Digital Karyotyping (MSDK). The MSDK method can be adapted to establish a test genomic methylation profile for a test cell of interest. By comparing the test profile to control profiles obtained with defined cells types, the test cell can be identified. The MSDK method can also be used to identify genes in a test cell (e.g., a cancer cell) the methylation of which is altered (increased or decreased) relative to a corresponding control cell (e.g., a normal cell of the same tissue as the cancer cell). This information provides the basis for methods for discriminating whether a test cell of interest (a) is the same as a control cell (e.g., a normal cell) or (b) is different from a control cell but is, for example, a pathologic cell such as a cancer cell. Such methods include, for example, assessing the level of DNA methylation or the level of expression of genes of interest, or the level of DNA methylation in a particular chromosomal area in test cells and comparing the results to those obtained with control cells.

More specifically, the invention features a method of making a methylation specific digital karyotyping (MSDK) library. The method includes: providing all or part of the genomic DNA of a test cell; exposing the DNA to a methylation-sensitive mapping restriction enzyme (MMRE) to generate a plurality of first fragments; conjugating to one terminus or to both termini of each of the first fragments a binding moiety, the binding moiety comprising a first member of an affinity pair, the conjugating resulting in a plurality of second fragments; exposing the plurality of second fragments to a fragmenting restriction enzyme

(FRE) to generate a plurality of third fragments, each third fragment containing at one terminus the first member of the affinity pair and at the other terminus the 5' cut sequence of the FRE or the 3' cut sequence of the FRE; contacting the plurality of third fragments with an insoluble substrate having bound thereto a plurality of second members of the affinity pair,the contacting resulting in a plurality of bound third fragments, each bound third fragment being a third fragment bound via the first and second members of the affinity pair to the insoluble substrate; conjugating to free termini of the bound third fragments a releasing moiety, the releasing moiety comprising a releasing restriction enzyme (RRE) recognition sequence and, 3' of the recognition sequence of the RRE, either the 5' cut sequence of the FRE or the 3' cut sequence of the FRE, the conjugating resulting in a plurality of bound fourth fragments, each bound fourth fragment (i) containing at one terminus the recognition sequence of the RRE and (ii) being bound via the first member of the affinity pair at the other terminus and the second member of the affinity pair to the insoluble substrate; and exposing the bound fourth fragments to the RRE, the exposing resulting in the release from the insoluble substrate of a MSDK library, the library comprising a plurality of fifth fragments, each fifth fragment comprising the releasing moiety and a MSDK tag, the tag consisting of a plurality of base pairs of the genomic DNA. Thus, the method results in the production of a plurality of MSDK tags.

In the method, the MMRE can be, e.g., Ascl, the FRE can be, e.g., NIaIII, and the RRE can be, e.g., Mrnel. The binding moiety can further include a 5' or 3' cut sequence of the MMRE. The binding moiety can also further include, between the 5' or 3' recognition sequence of the MMRE and the first member of an affinity pair, a linker nucleic acid sequence comprising a plurality of base pairs. The releasing moiety can further include, 5' of the RRE recognition sequence, an extender nucleic acid sequence comprising a plurality of base pairs. The test cell can be a vertebrate cell and the vertebrate test cell can be a mammalian test cell, e.g., a human test cell. Moreover the test cell can be a normal cell or, for example, a cancer cell, e.g., a breast cancer cell. The first member of the affinity pair can be biotin, iminobiotin, avidin or a functional fragment of avidin, an antigen, a haptenic determinant, a single-stranded nucleotide sequence, a hormone, a ligand for adhesion receptor, a receptor for an adhesion ligand, a ligand for a lectin, a lectin, a molecule containing all or part of an immunoglobulin Fc region, bacterial protein A, or bacterial protein G. The insoluble substrate can include, or be, magnetic beads.

Also provided by the invention is a method of analyzing a MSDK library. The method includes: providing a MSDK library made by the above-described method; and identifying the nucleotide sequences of one tag, a plurality of tags, or all of the tags. Identifying the nucleotide sequences of a plurality of tags can involve: making a plurality of ditags, each ditag containing two fifth fragments ligated together; forming a concatamer containing a plurality of ditags or ditag fragments, wherein each ditag fragment contains two MSDK tags; determining the nucleotide sequence of the concatamer; and deducing, from the nucleotide sequence of the concatamer, the nucleotide sequences of one or more of the MSDK tags that the concatamer contains. The ditag fragments can be made by exposing the ditags to the FRE. The method can further include, after making a plurality of ditags and prior to forming the concatamers, the number (abundance) of individual ditags is increased by PCR. The method can further include determining the relative frequency of some or all of the tags.

Another aspect of the invention is an additional method of analyzing a MSDK library. The method includes: providing a MSDK library made by the above-described method; identifying a chromosomal site corresponding to the sequence of a tag selected from the library. The method can further involve determining a chromosomal location, in the genome of the test cell, of an unmethylated full recognition sequence of the MMRE closest to the identified chromosomal site. These two steps can be repeated with a plurality of tags obtained from the library in order to determine the chromosomal location of a plurality of unmethylated recognition sequences of the MMRE. The identification of the chromosomal site and the determination of the chromosomal location can be performed by a process that includes comparing the nucleotide sequence of the selected tag to a virtual tag library generated using the nucleotide sequence of the genome or the part of a genome, the nucleotide sequence of the full recognition sequence of the MMRE, the nucleotide sequence of the full recognition sequence of the FRE, and the number of nucleotides separating the full recognition sequence of the RRE from the RRE cutting site.

In another aspect, the invention provides a method of classifying a biological cell. The method includes: (a) identifying the nucleotide sequences of one tag, a plurality of tags, or all of the tags in an MSDK library made as described above and determining the relative frequency of some or all of the tags, thereby obtaining a test MSDK profile for the test cell; (b) comparing the test MSDK profile to separate control MSDK expression profiles for one or more control cell types; (c) selecting a control MSDK profile that most closely resembles the test MSKD profile; and (d) assigning to the test cell a cell type that matches the cell type of the control MSDK profile selected in step (c). The test and control cells can be vertebrate cells, e.g., mammalian cells such as human cells. The control cell types can include a control normal cell and a control cancer cell of the same tissue as the normal cell. The control normal cell and the control cancer cell can be breast cells or of a tissue selected from colon, lung, prostate, and pancreas. The test cell can be a breast cell or of a tissue selected from of colon, lung, prostate, and pancreas. The control cell types can include cells of different categories of a cancer of a single tissue and the different categories of a cancer of a single tissue can include, for example, a breast ductal carcinoma in situ (DCIS) cell and an invasive breast cancer cell. The different categories of a cancer of a single tissue can alternatively include, for example, two or more of: a high grade DCIS cell, an intermediate grade DCIS cell; and a low grade DCIS cell. The control cell types can include two or more of: a lung cancer cell; a breast cancer cell; a colon cancer cell; a prostate cancer cell; and a pancreatic cancer. In addition, the control cell types can include an epithelial cell obtained from non-cancerous tissue and a myoepithelial cell obtained from non-cancerous tissue. Furthermore, the control cells can also include stem cells and differentiated cells derived therefrom (e.g., epithelial cells or myoepithelial cells) of the same tissue type. The control stem and differentiated cells therefrom can be of breast tissue, or of a tissue selected from colon, lung, prostate, and pancreas. The control stem and differentiated cells derived therefrom can be normal or cancer cells (e.g., breast cancer cells) or obtained from a cancerous tissue (e.g., breast cancer).

Another embodiment of the invention is a method of diagnosis. The method includes: (a) providing a test breast epithelial cell; (b) determining the degree of methylation of one or more C residues in a DNA sequence (e.g., in a gene) in the test cell, wherein the DNA (e.g., the gene) is selected from the Ascl sites identified by the MSDK tags listed in Table 5, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control epithelial cell obtained from non-cancerous breast tissue, wherein an altered degree of methylation of the one or more C residues in the test epithelial cell compared to the control epithelial cell is an indication that the test epithelial cell is a cancer cell. The altered degree of methylation can be a lower degree of methylation or a higher degree of methylation. The altered degree of methylation can be in the promoter region of the gene, an exon of the gene, an intron of the gene, or a region outside of the gene (e.g., in an intergenic region). The gene can be, for example, PRDM14 or ZCCHC 14.

The invention provides another method of diagnosis. The method includes: (a) providing a test colon epithelial cell; (b) determining the degree of methylation of one or more C residues in a DNA sequence (e.g., in a gene) in the test cell, wherein the DNA sequence (e.g., the gene) is selected from those identified by the MSDK tags listed in Table 2, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control epithelial cell obtained from non-cancerous colon tissue, wherein an altered degree of methylation of the one or more C residues in the test epithelial cell compared to the control epithelial cell is an indication that the test epithelial cell is a cancer cell. The altered degree of methylation can be a lower degree of methylation or a higher degree of methylation. In addition, the altered degree of methylation can be in the promoter region of the gene, an exon of the gene, an intron of the gene, or a region outside of the gene

(e.g., an intergenic region). The gene can be, for example, LHX3, TCF7L1, or LMX-IA.

Another method of diagnosis featured by the invention involves: (a) providing a test myoepithelial cell obtained from a test breast tissue; (b) determining the degree of methylation of one or more C residues in a DNA sequence (e.g., in a gene) in the test cell, wherein the DNA sequence (e.g., the gene) is selected from those identified by the

MSDK tags listed in Table 10, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control myoepithelial cell obtained from non-cancerous breast tissue, wherein an altered degree of methylation of the one or more C residues in the test myoepithelial cell compared to the control myoepithelial cell is an indication that the test breast tissue is cancerous tissue. The altered degree of methylation can be a lower degree of methylation or a higher degree of methylation. In addition, the altered degree of methylation can be in the promoter region of the gene, an exon of the gene, an intron of the gene, or a region outside of the gene (e.g., an intergenic region). The gene is can be, for example, HOXD4, SLC9A3R1, or CDC42EP5. Yet another method of diagnosis embodied by the invention involves: (a) providing a test fibroblast obtained from a test breast tissue; (b) determining the degree of methylation of one or more C residues in a DNA sequence (e.g., in a gene) in the test cell, wherein the DNA sequence (e.g., the gene) is selected from those identified by the MSDK tags listed in Tables 7 and 8, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control fibroblast obtained from non-cancerous breast tissue, wherein an altered degree of methylation of the one or more C residues in the test fibroblast compared to the control fibroblast is an indication that the test breast tissue is cancerous tissue. The altered degree of methylation can be a lower degree of methylation or a higher degree of methylation. In addition, the altered degree of methylation can be in the promoter region of the gene, an exon of the gene, an intron of the gene, or a region outside of the gene (e.g., an intergenic region). The gene can be, for example, Cxorfl2. In another aspect, the invention includes a method of determining the likelihood of a cell being an epithelial cell or a myoepithelial cell. The method involves: (a) providing a test cell; (b) determining the degree of methylation of one or more C residues in a DNA sequence (e.g., in a gene) in the test cell, wherein the DNA sequence (e.g., the gene) is selected from those identified by the MSDK tags listed in Table 12, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control myoepithelial cell and to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control epithelial cell, wherein the test cell is: (i) more likely to be a myoepithelial cell if the degree of methylation in the test sample more closely resembles the degree of methylation in the control myoepithelial cell; or (ii) more likely to be an epithelial cell if the degree of methylation in the test sample more closely resembles the degree of methylation in the control epithelial cell. The C residues can be in the promoter region of the gene, an exon of the gene, an intron of the gene, or in a region outside of the gene (e.g., an intergenic region). The gene can be, for example, LOC389333 or CDC42EP5. In another aspect, the invention includes a method of determining the likelihood of a cell being a stem cell, an differentiated luminal epithelial cell or a myoepithelial cell. The method involves: (a) providing a test cell; (b) determining the degree of methylation of one or more C residues in a DNA sequence (e.g., in a gene) in the test cell, wherein the DNA sequence (e.g., the gene) is selected from those identified by the MSDK tags listed in Table 15 or 16, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control stem cell, to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control differentiated luminal epithelial cell, and to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control myoepithelial cell, wherein the test cell is: (i) more likely to be a stem cell if the degree of methylation in the test sample more closely resembles the degree of methylation in the control stem cell; (ii) more likely to be a differentiated luminal epithelial cell if the degree of methylation in the test sample more closely resembles the degree of methylation in the control epithelial cell; or (iii) more likely to be a myoepithelial cell if the degree of methylation in the test sample more closely resembles the degree of methylation in the control myoepithelial cell. The C residues can be in the promoter region of the gene, an exon of the gene, an intron of the gene, or in a region outside of the gene (e.g., an intergenic region). The gene can be, for example, SOX13, SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and HOXAlO.

The invention also features a method of diagnosis that involves: (a) providing a test cell from a test tissue; (b) determining the degree of methylation of one or more C residues in a PRDM14 gene in the test cell, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in the PRDM 14 gene in a control cell obtained from non-cancerous tissue of the same tissue as the test cell, wherein an altered degree of methylation of the one or more C residues in the test cell compared to the control cell is an indication that the test cell is a cancer cell. The altered degree of methylation can be a lower degree of methylation or a higher degree of methylation. In addition, the altered degree of methylation can be in the promoter region of the gene, an exon of the gene, an intron of the gene, or a region outside of the gene (e.g., an intergenic region). The test and control cells can be breast cells or of a tissue selected from colon, lung, prostate, and pancreas.

Another embodiment of the invention is a method of diagnosis that includes: (a) providing a test sample of breast tissue comprising a test epithelial cell; (b) determining the level of expression in the test epithelial cell of a gene selected from those listed in Table 5, wherein the gene is one that is expressed in a breast cancer epithelial cell at a substantially altered level compared to a compared to a normal breast epithelial cell; and (c) classifying the test cell as: (i) a normal breast epithelial cell if the level of expression of the gene in the test cell is not substantially altered compared to a control level of expression for a normal breast epithelial cell; or (ii) a breast cancer epithelial cell if the level of expression of the gene in the test cell is substantially altered compared to a control level of expression for a normal breast epithelial cell. The gene is can be, for example, PRDM 14 or ZCCHC 14. The alteration in the level of expression can be an increase in the level of expression or a decrease in the level of expression.

Another aspect of the invention is a method of diagnosis that includes:

(a) providing a test sample of colon tissue comprising a test epithelial cell;

(b) determining the level of expression in the test epithelial cell of a gene selected from those listed in Table 2, wherein the gene is one that is expressed in a colon cancer epithelial cell at a substantially altered level compared to a compared to a normal colon epithelial cell; and (c) classifying the test cell as: (i) a normal colon epithelial cell if the level of expression of the gene in the test cell is not substantially altered compared to a control level of expression for a normal colon epithelial cell; or (ii) a colon cancer epithelial cell if the level of expression of the gene in the test cell is substantially altered compared to a control level of expression for a normal colon epithelial cell. The gene can be, for example, LHX3, TCF7L1, or LMX-IA. The alteration in the level of expression can be an increase in the level of expression or a decrease in the level of expression.

Another method of diagnosis included in the invention involves: (a) providing a test sample of breast tissue comprising a test stromal cell; (b) determining the level of expression in the stromal cell of a gene selected from those listed in Tables 7, 8, and 10, wherein the gene is one that is expressed in a cell of the same type as the test stromal cell at a substantially altered level when present in breast cancer tissue than when present in normal breast tissue; and (c) classifying the test sample as: (i) normal breast tissue if the level of expression of the gene in the test stromal cell is not substantially altered compared to a control level of expression for a control cell of the same type as the test stromal cell in normal breast tissue; or (ii) breast cancer tissue if the level of expression of the gene in the test stromal cell is substantially altered compared to a control level of expression for a control cell of the same type as the test stromal cell in normal breast tissue. The test and control stromal cells can be myoepithelial cells and the genes can be those listed in Table 10, e.g., HOXD4, SLC9A3R1, or CDC32EP5. Alternatively, the test and control stromal cells can be fibroblasts and the genes can be those listed in

Tables 7 and 8, e.g., Cxorfl. The alteration in the level of expression can be an increase in the level of expression or a decrease in the level of expression.

In another aspect, the invention includes a method of determining the likelihood of a cell being an epithelial cell or a myoepithelial cell. The method includes: (a) providing a test cell; (b) determining the level of expression in the test sample of a gene selected from the group consisting of those identified by the MSDK tags listed in Table 12; (c) determining whether the level of expression of the selected gene in the test sample more closely resembles the level of expression of the selected gene in (i) a control myoepithelial cell or (ii) a control epithelial cell; and (d) classifying the test cell as: (i) likely to be a myoepithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control myoepithelial cell; or (ii) likely to be an epithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control epithelial cell. The gene can be, for example, LOC389333 or CDC42EP5. In another aspect, the invention includes a method of determining the likelihood of a cell being a stem cell, a differentiated lumnial epithelial cell, or a myoepithelial cell. The method includes: (a) providing a test cell; (b) determining the level of expression in the test sample of a gene selected from the group consisting of those identified by the MSDK tags listed in Table 15 or 16; (c) determining whether the level of expression of the selected gene in the test sample more closely resembles the level of expression of the selected gene in (i) a control stem cell, (ii) a control differentiated luminal epithelial cell, or (iii) a control myoepithelial cell; and (d) classifying the test cell as: (i) likely to be a stem cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control stem cell; (ii) likely to be an differentiated luminal epithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control differentiated luminal epithelial cell, or (iii) likely to be a myoepithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control myoepithelial cell. The gene can be, for example, SOX13, SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and HOXAlO. Also embodied by the invention is a method of diagnosis that includes:

(a) providing a test cell; (b) determining the level of expression in the test cell of a PRDMl 4 gene; and (c) classifying the test cell as: (i) a normal cell if the level of expression of the gene in the test cell is not substantially altered compared to a control level of expression for a control normal cell of the same tissue as the test cell; or (ii) a cancer cell if the level of expression of the gene in the test cell is substantially altered compared to a control level of expression for a control normal cell of the same tissue as the test cell. The alteration in the level of expression can be an increase in the level of expression or a decrease in the level of expression. The test and control cells can be breast cells or of a tissue selected from colon, lung, prostate, and pancreas. The invention also provides a single stranded nucleic acid probe that includes:

(a) the nucleotide sequence of a tag selected from those listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16; (b) the complement of the nucleotide sequence; or (c) the Ascl sites defined by the MSDK tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16.

In another aspect, there is provided an array containing a substrate having at least 10, 25, 50, 100, 200, 500, or 1,000 addresses, wherein each address has disposed thereon a capture probe that includes: (a) a nucleic acid sequence consisting of a tag nucleotide sequence selected from those listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16; (b) the complement of the nucleic acid sequence; or (c) the Ascl sites defined by the MSDK tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16. The invention also features a kit comprising at least 10, 25, 50, 100, 200, 500, or

1,000 probes, each probe containing: (a) a nucleic acid sequence comprising a tag nucleotide sequence selected from those listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16; (b) the complement of the nucleic acid sequence; (c) the Ascl sites defined by the MSDK tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16.

Another aspect of the invention is kit containing at least 10, 25, 50, 100, 200, 500, or 1,000 antibodies each of which is specific for a different protein encoded by a gene identified by a tag selected from the group consisting of the tags listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16.

As used herein, an "affinity pair" is any pair of molecules that have an intrinsic ability to bind to each other. Thus, affinity pairs include, without limitation, any receptor/ligand pair, e.g., vitamins (e.g., biotin)/vitamin-binding proteins (e.g., avidin or streptavidin); cytokines (e.g., interleukin-2)/cytokine receptors (e.g., interleukin-2); hormones (e.g., steroid hormones)/hormone receptors (e.g., steroid hormone receptors); signal transduction ligands/signal transduction receptors; adhesion ligands/adhesion receptors; death domain molecule-binding ligands/death domain molecules; lectins (e.g., pokeweed mitogen, pea lectin, concanavalin A, lentil lectin, phytohemagglutinin (PHA) from Phaseolus vulgaris, peanut agglutinin, soybean agglutinin, Ulex europaeus agglutinin-I, Dolichos biflorus agglutinin, Vicia villosa agglutinin and Sophora japonica agglutinin/lectin receptors (e.g., carbohydrate lectin receptors); antigens or haptens (e.g., trinitrophenol or biotin)/antibodies (e.g., antibody specific for trinitrophenol or biotin); immunoglobulin Fc fragments/immunoglobulin Fc fragment binding proteins (e.g., bacterial protein A or protein G). Ligands can serve as first or second members of an affinity pair, as can receptors. Where a ligand is used as the first member of the affinity pair the corresponding receptor is used as the second member of the affinity pair and where a receptor is used as the first member of the affinity pair, the corresponding receptor is used as the second member of the affinity pair. Functional fragments of polypeptide first and second members of affinity pairs are fragments of the full-length, mature first or second members that are shorter than the full-length, mature first or second members but have at least 25% (e.g., at least: 30%; 40%; 50%; 60%; 70%; 80%; 90%; 95%; 98%; 99%; 99.5%; 100%; or even more) of the ability of the full-length, mature first or second members to bind to corresponding second or first members, respectively. The nucleotide sequences of all the identified genes in Tables 2, 5, 7, 8, 10, 12, 15 and 16 are available on public genetic databases (e.g., GeneBank). These sequences are incorporated herein by reference.

As used herein, a "substantially altered" level of expression of a gene in a first cell (or first tissue) compared to a second cell (or second tissue) is an at least 2-fold (e.g., at least: 2-; 3-; A-; 5-; 6~; 7-; 8-; 9-; 10-; 15-; 20-; 30-; 40-; 50-; 75-; 100-; 200-; 500-; 1,000-; 2000-; 5,000-; or 10,000-fold) altered level of expression of the gene. It is understood that the alteration can be an increase or a decrease.

As used herein, breast "stromal cells" are breast cells other than epithelial cells. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. In case of conflict, the present document, including definitions, will control. Preferred methods and materials are described below, although methods and materials similar or equivalent to those described herein can be used in the practice or testing of the present invention. All publications, patent applications, patents and other references mentioned herein are incorporated by reference in their entirety. The materials, methods, and examples disclosed herein are illustrative only and not intended to be limiting.

Other features and advantages of the invention, e.g., assessing the methylation of an entire genome, will be apparent from the following description, from the drawings and from the claims.

DESCRIPTION OF DRAWINGS

Fig. 1 is a diagrammatic representation of the generation of a restriction enzyme 5' cut sequence and 3 ' cut sequence by the restriction enzyme cutting DNA at the restriction enzyme's recognition sequence, m the diagram are shown the two strands of a segment of double stranded DNA containing a restriction enzyme recognition sequence in which each of the nucleotides constituting the recognition sequence are shown as an N. The exemplary restriction enzyme recognition sequence in the diagram is a six base pair recognition sequence and cutting by the particular restriction enzyme results in a 3 ' two nucleotide overhang. The N-containing sequences constituting the restriction enzyme recognition sequence and the restriction enzyme's 3' and 5' cut sequences are boxed and appropriately labeled. Those skilled in the art will appreciate that 5' and 3' termini generated by the multiple restriction enzymes available differ greatly (in nucleotide content, whether cohesive termini are generated, and, if they are, in the nature and number of nucleotides in the overhang). Nevertheless, in the sense that all termini (5' and 3' cut sequences) produced by the action of restriction enzymes that cut at their recognition sequences consist of nucleotides derived from the relevant restriction enzyme recognition sequence, 5' and 3' restriction enzyme cut sequences share qualitative features and differ only in how these nucleotides are distributed between the 5' and 3' cut sequences.

Fig. 2 is a schematic depiction of the MSDK procedure described in Examples 1 and 2.

Figs. 3 -5 are diagrammatic representations of the results of a methylation- detecting sequence analysis of segments of the LHX3 gene region (Fig. 3; SEQ ID

NO:3), the LMX-IA gene region (Fig. 4; SEQ ID NO: 5), and the TCF7L1 gene region (Fig. 5; SEQ ID NO:4) shown in Figs. 6-8, respectively. The circles represent potential methylation sites (CpG) in the analyzed segment of SEQ ID NOs:3, 5, and 4. The order of circles (starting from the left of the rows of circles) is that of the CpG dinucleotides in the analyzed segments of SEQ ID NOs: 3, 5 and 4 (starting from the 5' end of the analyzed segment nucleotide sequences). The analyses were performed on DNA from wild-type HCTl 16 human colon cancer cells ("WT") and HCTl 16 cells having both alleles of their DNTMl and DNMT3b methyltransferase genes "knocked out" ("DKO"). Each circle is pie chart with the amount of shading indicating the frequency (0% -100%) at which the relevant potential methylation site was found to be methylated. The top lines under the circles are linear depictions of the relevant gene transcripts and include the exons (shaded boxes) and introns (lines between the shaded boxes) and the bottom line under the circles are linear depictions of the chromosome on which the genes are located. On the chromosome depictions are shown the locations of the MSDK tag sequences that indicated the locations of the relevant Ascl recognition sequences, which locations are also shown. The numbering on the bottom lines indicates the base pair (bp) numbers on the chromosomes and the numbering on the top lines indicate the bp numbers, in the chromosomes, of the transcription start sites and termination sites. The transcription initiation sites and the directions of transcription are also shown.

Fig. 6 A is a depiction of the nucleotide sequence (SEQ ID NO: 3) of a region of the LHX3 gene containing the MSDK tag sequence (bold and underlined) that identified the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO: 3 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3 ' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3 ' end of the reverse PCR primer target sequence

(shown in italics and underlined). The sequenced segment spans bp -196 to bp +172 (relative to the LHX3 gene transcription initiation site) and thus the last 23 CpG in the sequenced segment are within the promoter region and the first 26 CpG are in exon 1. Fig. 6B is a depiction of the nucleotide sequence (SEQ ID NO:1545) of a region of the LHX3 gene within SEQ ID NO: 3 containing the relevant Ascl site (bold and underlined) and multiple CpG dinucleotides (shaded).

Fig. 7 A is a depiction of the nucleotide sequence (SEQ ID NO: 5) of a region of the LMX-IA gene containing the MSDK tag sequence (bold and underlined) that identified the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO:5 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp -842 to bp -609 (relative to the LMX-IA gene transcription initiation site) and thus the whole of the sequenced segment is within the promoter region.

Fig. 7B is a depiction of the nucleotide sequence (SEQ ID NO: 1546) of a region of the LMX-IA gene within SEQ ID NO: 5 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). Fig. 8A is a depiction of the nucleotide sequence (SEQ ID NO:4) of a region of the TCF7L1 gene containing the MSDK tag sequence (bold and underlined) that identified the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO:4 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3 ' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp +782 to bp +1003 (relative to the TCF7L1 gene transcription initiation site) and thus the first six CpG in the sequenced segment are within exon 1 and the last 19 CpG are in intron 3-4. Fig. 8B is a depiction of the nucleotide sequence (SEQ ID NO: 1547) of a region of the TCF7L1 gene within SEQ ID NO:4 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). .

Figs. 9 - 15 are diagrammatic representations of the results of a methylation-detecting sequence analysis of the segments of, respectively, the PRDM14 gene region (Fig. 9; SEQ ID NO: 1), the ZCCHC14 gene region (Fig. 10; SEQ ID NO:2), the HOXD4 gene region (Fig. 11; SEQ ID NO:6), the SLC9A3R1 gene region (Fig. 12; SEQ ID NO:7), the LOC38933 gene region (Fig. 13; SEQ ID NO: 10), the CDC42EP5 gene region (Fig. 14; SEQ ID NO:8), and the Cxorfl2 gene region (Fig. 15; SEQ ID NO:9) shown in Figs.l6A-22A, respectively. The circles represent potential methylation sites (CpG) in the analyzed segments. The order of circles (starting from the left of the rows of circles) is that of the CpG dinucleotides in the analyzed segments (starting from the 5' end of the analyzed segment nucleotide sequences). The analyses were performed on DNA from the indicated cell obtained from the indicated samples (see Table 3). Samples used for the generation of MSDK libraries are marked with an asterisk. Each circle is a pie chart with the amount of shading indicating the frequency (0% - 100%) at which the relevant potential methylation site was found to be methylated. The top (bold) lines under the circles are linear depictions of the relevant gene transcripts and include the exons (shaded boxes) and introns (lines between the shaded boxes) and the bottom lines under the circles are linear depictions of the chromosomes on which the genes are located. On the chromosome depictions are shown the locations of the MSDK tag sequences that indicated the location of the relevant Ascl recognition sequences, which locations are also shown. The numbering on the bottom lines indicates the bp numbers for the chromosomes and the numbering on the top lines indicate the bp numbers, in the chromosomes, of the transcription start sites and termination sites. The transcription initiation sites and the directions of transcription are also shown. Fig. 15 provides the above-listed information for the HCFCl gene as well as the Cxorfl2 gene. As can be seen for the figure, the two genes are located relatively close together on the X chromosome.

Fig. 16A is a depiction of the nucleotide sequence (SEQ ID NO:1) of a region of the PRDM 14 gene containing the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO: 1 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp +666 to bp +839 (relative to the PRDM14 gene transcription initiation site) and thus the whole sequenced segment is within intron 1-2.

Fig. 16B is a depiction of the nucleotide sequence (SEQ ID NO: 1548) of a region of the PRDM14 gene within SEQ ID NO: 1 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). Fig. 17A is a depiction of the nucleotide sequence (SEQ ID NO:2) of a region of the ZCCHC 14 gene containing the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO: 2 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp +79 to bp +292 (relative to the ZCCHC 14 gene transcription initiation site) and thus the last 14 CpG in the sequenced segment are within exon 1 and the first 7 CpG are in intron 1-2. Fig. 17B is a depiction of the nucleotide sequence (SEQ ID NO: 1549) of a region

) of the ZCCHC14 gene within SEQ ID NO:2 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded).

Fig. 18Ais a depiction of the nucleotide sequence (SEQ ID N0:6) of a region of the H0XD4 gene containing the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO:6 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp +986 to bp +1,189 (relative to the H0XD4 gene transcription initiation site) and thus the whole sequenced segment is within intron 1-2.

Fig. 18B is a depiction of the nucleotide sequence (SEQ ID NO: 1550) of a region of the HOXD4 gene within SEQ ID NO: 6 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded).

Fig. 19Ais a depiction of the nucleotide sequence (SEQ ID NO:7) of a region of the SLC9A3R1 gene containing the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO:7 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp +11,713 to bp +11,978 (relative to the SLC9A3R1 gene transcription initiation site) and thus the whole sequenced segment is within intron 1-2. Fig. 19B is a depiction of the nucleotide sequence (SEQ ID NO:1551) of a region of the SLC9A3R1 gene within SEQ ID NO:7 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded).

Fig. 2OA is a depiction of the nucleotide sequence (SEQ ID NO: 10) of a region of the LOC389333 gene containing the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID

NO: 10 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp +518 to bp +762 (relative to the LOC389333 gene transcription initiation site) and thus the last 10 CpG in the sequenced segment are within exon 1 and the first 21 CpG are within intron 1-2.

Fig. 2OB is a depiction of the nucleotide sequence (SEQ ID NO: 1552) of a region of the LOC389333 gene within SEQ ID NO: 10 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). Fig. 21 A is a depiction of the nucleotide sequence (SEQ ID NO : 8) of a region of the CDC42EP5 gene containing the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO: 8 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp +7,991 to bp +8,193 (relative to the CDC42EP5 gene transcription initiation site) and thus the whole the sequenced segment is within exon 3.

Fig. 21B is a depiction of the nucleotide sequence (SEQ ID NO:1553) of a region of the CDC42EP5 gene within SEQ ID NO : 8 containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded).

Fig. 22A is a depiction of the nucleotide sequence (SEQ ID NO:9) of a region of the Cxorf 12 gene containing the MSDK tag sequence (bold and underlined) that identified the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). The segment of SEQ ID NO:9 subjected to methylation-detecting sequence analysis starts at the nucleotide after the 3' end of the forward PCR primer target sequence (shown in italics and underlined) used for the sequencing analysis and ends at the nucleotide before the 3' end of the reverse PCR primer target sequence (shown in italics and underlined). The sequenced segment spans bp -838 to bp -639 (relative to the Cxorf 12 gene transcription initiation site) and thus the whole sequenced segment is within the promoter region. Fig. 22B is a depiction of the nucleotide sequence (SEQ ID NO: 1555) of a region of the Cxorfl2 gene within SEQ ID NO:9 containing the MSDK tag sequence (bold and underlined) that identified the relevant Ascl recognition sequence (in capital letters and underlined) and multiple CpG dinucleotides (shaded). Figs. 23 A-F are a series of bar graphs showing the results of quantitative methylation specific PCR (qMSP) analyses of the PRDM14 (Fig. 23A), HOXD4 (Fig. 23B), SLC9A3R1 (Fig. 23C), CDC42EP5 (Fig. 23D), LOC389333 (Fig. 23E), and Cxorfl2 (Fig. 23F) genes in epithelial cells (left set of normal and tumor cell bars), myoepithelial cells (middle set of normal and tumor cell bars), and fibroblast-enriched stromal cells (right set of normal and tumor cells) isolated from the indicated normal breast tissue and breast carcinoma samples. The average Ct value for each gene was normalized against the ACTB value (see Example 1). The data ("Relative methylation (%)") are percentages relative to the ACTB value. Samples used for generation of MSDK libraries are indicated by asterisks. The PRDM14 gene is almost exclusively methylated in tumor epithelial cells and the LOC389333 gene is preferentially methylated in epithelial cells (both tumor and normal) compared to other cell types. The H0XD4, SLC9A3R1, and CDC42EP5 genes, besides being differentially methylated between normal and DCIS and myoepithelial cells, are also methylated in other cell types. The H0XD4 gene is differentially methylated between normal and tumor epithelial cells and frequently methylated in stromal fibroblasts, while the SLC9A3R1 and CDC43EP5 genes are frequently methylated in stromal fibroblasts and occasionally in epithelial cells. The Cxorfl2 gene is hypermethylated in tumor fibroblast enriched stromal cells compared to normal cells of the same type and is also methylated in a fraction of epithelial cells.

Fig. 24 is a bar graph showing the results of qMSP analyses of the PRDM 14 gene in a panel of normal breast tissues, benign breast tumors (fibroadenomas, papillomas, and fibrocystic disease), and breast carcinomas. The data were computed as described for Fig. 23. 500% was set as the upper limit of relative methylation although a few samples showed a difference above this threshold.

Figs. 25 A-D are a series of bar graphs showing the results of expression analyses of the PRDM14 (Fig. 25A), Cxorfl2 (Fig. 25B), CDC42EP5 (Fig. 25C), and HOXD4 (Fig. 25D) genes in normal breast and breast carcinoma (tumor) epithelial cells, fibroblast-enriched stromal cells (stroma), and myoepithelial cells and in invasive breast carcinoma cell myofibroblasts. The average Ct value for each gene was normalized against the RPL39 value (see Example 1). The data ("Relative expression (%)") are percentages relative to the RPL39 value. Using RPL 19 and RPS 13 values for normalization gave essentially the same results. The PRDM14 gene was relatively overexpressed in invasive breast carcinoma epithelial cells. The Corfl2 gene was expressed at a relatively higher level in normal than in tumor fibroblast-enriched stromal cells. The CDC42EP5 and HOXD4 genes showed higher expression in DCIS myoepithelial cells and invasive breast carcinoma myofibroblasts compared to normal myoepithelial cells and also, in the case of the CDC42EP5 gene, to normal epithelial cells.

Fig. 26A is a schematic representation of the procedure used for tissue fractionation and purification of the various cell types from normal breast tissue. Cells were captured by antibody-coupled magnetic beads as indicated by the figure. Fig. 26B is a is a series of photographs of ethidium bromide-stained electrophoretic gels of semi-quantitative RT-PCR analyses of selected genes from the purified cell fractions isolated from normal breast tissue. PPIA was used as a loading control. The triangles indicate an increasing number of PCR cycles (25, 30, and 35).

Fig. 26C is a series of graphs showing the ratio and location of statistically significant (p<0.05) tags, generated by MSDK, that are differentially methylated in different cell types isolated from normal mammary tissue. Dots corresponding to genes selected for further validation are circled. The X-axis represents the ratio of normalized tags from the indicated libraries in the various comparisons. CD44/A11 indicates the comparison of mammary stem cells (CD44+) against all differentiated cells (CD 10+, CD24+, and MUC 1+).

Fig. 27Ais a series of diagrammatic representations of the results of a methylation- detecting sequence analysis of segments of the SLC9A3R1 gene region, the FNDCl gene region, the FOXCl gene region, the PACAP gene region, the DDN gene region, the CDC42EP5 gene region, the LHXl gene region, the SOXl 3 gene region, and the DTX gene region. The circles represent potential methylation sites (CpG) in the analyzed semgment of SEQ ID NOs:7, 8, and 11-18. The order of the circles (starting from the left of the rows of circles) is that of the CpG dinucleotides in the analyzed segments of SEQ ID NOs:7, 8, and 11-18 (starting from the 5' end of the analyzed segment nucleotide sequences). The analyses were performed on DNA isolated from CD44+, CD24+, MUC 1+, and CD 10+ cell populations. Each circle is a pie chart with the amount of shading indicating the frequency (0-100%) at which the relevant potential methylation site was found to be methylated. The top lines under the circles are linear depictions of the relevant gene transcripts and include the exons (shaded boxes) and introns (lines between the shaded boxes) and the bottom line under the circles are linear depictions of the chromosome on which the genes are located. On the chromosome depictions are shown the locations of the MSDK tag sequences that indicated the locations of the relevant Ascl recognition sequences, which locations are also shown. The numbering on the bottom lines indicates the base pair (bp) numbers on the chromosomes and the numbering on the top lines indicate the bp numbers, in the chromosomes, of the transcription start sites and termination sites. The transcription initiation sites and the directions of transcription are also shown.

Fig. 27B is a series of bar graphs showing the results of quantitative methylation specific PCR (qMSP) analyses of the SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and HOXAlO genes in CD44+, CD10+, MUCH, and CD24+ cells populations from women of different ages (18-58 years old) and reproductive history. The average Ct value for each gene was normalized against the ACTB value. The data ("Relative expression (%)") are percentages relative to the RPL39 value.

Fig. 28 is a series of bar graphs showing the results of expression analyses of the SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and HOXAlO genes in CD44+, CD 10+, MUC 1+, and CD24+ cells isolated from normal breast tissue. The average Ct value for each gene was normalized against the RPL39 value. The data ("Relative expression (%)") are percentages relative to the RPL39 value.

Figs. 29A-29B are a series of bar graphs depicting the results of quantitative methylation specific PCR (qMSP) analyses of DNA from (A) the SLC9A3R1, FNDCl, FOXCl, PACA¹P, LHXl, and HOXAlO genes in putative breast cancer stem cells (T- EPCR+) and cells with more differentiated phenotype from the same tumor (T-CD24+), and (B) the HOXAlO, FOXCl, PACAP, and LHXl genes from matched primary tumors (indicated by a star) and distant metastases (DM) collected from different organs. The average Ct value for each gene was normalized against the RPL39 value (see Example 1). The data ("Relative expression (%)") are percentages relative to the RPL39 value.

Fig. 30 is a depiction of the nucleotide sequence (SEQ ID NO: 11) of a region of the FNDCl gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp -285 to bp - 614 (relative to the FNDCl gene transcription initiation site) and thus the whole sequenced segment is within the promoter region.

Fig. 31 is a depiction of the nucleotide sequence (SEQ ID NO: 12) of a region of the FOXCl gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp 5250 to bp 4976 (relative to the FOXCl gene transcription initiation site) and thus the whole sequenced segment is within the promoter region.

Fig. 32 is a depiction of the nucleotide sequence (SEQ ID NO: 13) of a region of the PACAP gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp 4404 to bp 4736 (relative to the PACAP gene transcription initiation site) and thus the whole sequenced segment is within the promoter region.

Fig. 33 is a depiction of the nucleotide sequence (SEQ ID NO: 14) of a region of the DDN gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp 2108 to bp 2290 (relative to the PACAP gene transcription initiation site) and thus the whole sequenced segment is within exon 2.

Fig. 34 is a depiction of the nucleotide sequence (SEQ ID NO:15) of a region of the LHXl gene containing the relevant A scl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp 3600 to bp 3810 (relative to the LHXl gene transcription initiation site) and thus the whole sequenced segment is within introns 3-4.

Fig. 35 is a depiction of the nucleotide sequence (SEQ ID NO: 16) of a region of the SOXl 3 gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp 669 to bp 374 (relative to the SOX 13 gene transcription initiation site) and thus the whole sequenced segment is within the promoter area.

Fig. 36 is a depiction of the nucleotide sequence (SEQ ID NO: 17) of a region of the

DTX gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp 228 to bp 551

(relative to the DTX gene transcription initiation site) and thus the whole sequenced segment is within the promoter area.

Fig. 37 is a depiction of the nucleotide sequence (SEQ ID NO: 18) of a region of the

HOXAlO gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp

4270 to bp 4634 (relative to the HOXAlO gene transcription initiation site) and thus the whole sequenced segment is within the promoter area.

Fig. 38 is a depiction of the nucleotide sequence (SEQ ID NO:1543) of a region of the SLC9A3R1 gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp

11713 to bp 11978 (relative to the SLC9A3R1 gene transcription initiation site) and thus the whole sequenced segment is within introns 1-2.

Fig. 39 is a depiction of the nucleotide sequence (SEQ ID NO:11544) of a region of the CDC42Ep5 gene containing the relevant Ascl recognition sequence (in bold and underlined) and multiple CpG dinucleotides (shaded). The sequenced segment spans bp

7855 to bp 8058 (relative to the CDC42Ep5 gene transcription initiation site) and thus the whole sequenced segment is within exon 3.

DETAILED DESCRIPTION

Various aspects of the invention are described below.

Methylation Specific Digital Karyotyping (^"MSDK)

MSDK is a method of assessing the relative level of methylation of an entire genome, or part of a genome, of a cell of interest. The cell can be any DNA-containing biological cell in which the DNA is subject to methylation, e.g., prokaryotic cells (e.g., bacteria) or eukaryotic cells (e.g., yeast cells, protozoan cells, invertebrate cells, or vertebrate (e.g., mammalian) cells).

Vertebrate cells can be from any vertebrate species, e.g., reptiles (e.g., snakes, alligators, and lizards), amphibians (e.g., frogs and toads), fish (e.g., salmon, sharks, or trout), birds (e.g., chickens, turkeys, eagles, or ostriches), or mammals. Mammals include, for example, humans, non-human primates (e.g., monkeys, baboons, or chimpanzees), horses, bovine animals (e.g., cows, oxen, or bulls), whales, dolphins, porpoises, pigs, sheep, goats, cats, dogs, rabbits, gerbils, guinea pigs, hamsters, rats, or mice. Vertebrate and mammalian cells can be any nucleated cell of interest, e.g., epithelial cells (e.g., keratinocytes), myoepithelial cells, endothelial cells, fibroblasts, melanococytes, hematological cells (e.g., macrophages, monocytes, granulocytes, T lymphocytes (e.g., CD4+ and CD8+ lymphocytes), B-lymphocytes, natural killer (NK) cells, interdigitating dendritic cells), nerve cells (e.g., neurons, Schwann cells, glial cells, astrocytes, or oligodendrocytes), muscle cells (smooth and striated muscle cells), chondrocytes, osteocytes. Also of interest are stem cells, progenitor cells, and precursor cells of any of the above-listed cells. Moreover the method can be applied to malignant forms of any of cells listed herein.

The cells can be of any tissue or organ, e.g., skin, eye, peripheral nervous system (PNS; e.g., vagal nerve), central nervous system (CNS; e.g., brain or spinal cord), skeletal muscle, heart, arteries, veins, lymphatic vessels, breast, lung, spleen, liver, pancreas, lymph node, bone, cartilage, joints, tendons, ligaments, gastrointestinal tissue (e.g., mouth, esophagus, stomach, small intestine, large intestine (e.g., colon or rectum)), genitourinary system (e.g., kidney, bladder, uterus, vagina, ovary, ureter, urethra, prostate, penis, testis, or scrotum). Cancer cells can be of any of these organs and tissues and include, without limitation, breast cancers (any of the types and grades recited herein), colon cancer, prostate cancer, lung cancer, pancreatic cancer, melanoma. MSDK can be performed on an entire genome of a cell, e.g., whole DNA extracted from an entire cell or the nucleus of a cell. Alternatively, it can be earned out on part of a cell, e.g., by extracting DNA from mutant cells lacking part of a genome, chromosome microdissection, or subtractive/differential hybridization. The method is performed on double-stranded DNA and, unless otherwise stated, in describing MSDK, the term "DNA" refers to double-stranded DNA.

Method ofmakins a MSDK library In the first step of the MSDK, genomic DNA is exposed to a methylation- sensitive mapping restriction enzyme (MMRE) that cuts the DNA at sites having the recognition sequence for the relevant MMRE. The MMRE can be any MMRE. In eukaryotic cells, methylation generally occurs at C nucleotides in CpG dinucleotide sequences in DNA. The term "CpG" refers to dinucleotide sequences that occur in DNA and consist of a C nucleotide and G nucleotide immediately 3' of the C nucleotide. The "p" in "CpG" denotes the phosphate group that occurs between the C and G nucleoside residues in the CpG dinucleotide sequence.

The MMRE recognition sequence can contain one, two, three, or four C residues that are susceptible to methylation. If one (or more) of the C residues in a MMRE recognition sequence is methylated, the MMRE does not cut the DNA at the relevant MMRE recognition sequence Examples of useful MMRE include, without limitation, Ascl, Aatll, AcU, Afel, Agel, AsisIAval, BceAI, BssHI, CM, Eagl, Hpy99I, MIuI, Narl, Notl, SacII, or ZraAI. The Ascl recognition sequence is GGCGCGCC and thus contains two methylation sites (CpG sequences). If either one or both is methylated, the recognition site is not cut by Ascl. There are approximately 5,000 Ascl recognition sites per human genome.

Exposure of the genomic DNA to the MMRE results in a plurality of first fragments, the absolute number of which will depend on the relative number of MMRE recognition sites that are methylated. The more that are methylated, the fewer first fragments will result. Most of the first fragments will have at one terminus the MMRE 5' cut sequence (see definition below) and at the other terminus the MMRE 3' cut sequence (see definition below). For each chromosome, two fragments with MMRE cut sequences at only one terminus will be generated; these first fragments are referred to herein as terminal first fragments. One such terminal first fragment contains the 5' terminus of the chromosome at one end and a MMRE 3' cut sequence at the other end and the other terminal fragment contains the 3' terminus of the chromosome at one end and a MMRE 5' cut sequence at the other end.

As used herein, a "5' cut sequence" of a restriction enzyme that cuts DNA within the restriction enzyme's recognition sequence is the portion of the restriction enzyme's recognition sequence at the 5' end of a fragment containing the 3' end of the restriction enzyme recognition sequence that is generated by cutting of DNA by the restriction enzyme. As used herein, a "3' cut sequence" of a restriction enzyme that cuts DNA within the restriction enzyme's recognition sequence is the portion of the restriction enzyme's recognition sequence at the 3' end of a fragment containing the 5' end of the restriction enzyme recognition sequence that is generated by cutting of DNA by the restriction enzyme. 5' and 3' cut restriction enzyme cut sequences are illustrated in Fig. 1.

To the termini of the first fragments are conjugated a first member of an affinity pair (see definition in Summary section), e.g., biotin or iminobiotin. This can be achieved by, for example, ligating to the MMRE 5' and 3' cut sequence-containing termini a binding moiety. The binding moiety contains the first member of the affinity pair conjugated (e.g., by a covalent bond or any other stable chemical linkage, e.g., a coordination bond, that can withstand the relatively mild chemical conditions of the MSDK methodology) to either a MMRE 5' cut sequence or a MMRE 3' cut sequence. The majority of the fragments (referred to herein as second fragments) resulting from attachment by this method of the first members of the affinity pair will have first members of an affinity pair bound to both their termini. Second fragments resulting from terminal first fragments will of course have first members of the affinity pair only at one terminus, i.e., the terminus containing the MMRE cut sequence.

The binding moiety can, optionally, also contain a linker (or spacer) nucleotide sequence of any convenient length, e.g., one to 100 base pairs (bp), three to 80 bp, five to 70 bp, seven to 60 bp, nine to 50, or 10 to 40 bp. The linker (or spacer) can be, for example, 30, 31, 32, 33, 34, 35, 26, 37, 38, or 40 bp long. As will be apparent, the linker must not include a fragmenting restriction enzyme (see below) recognition sequence.

Instead of using the above-described binding moiety to attach the first members of an affinity pair to the termini of first fragments, the attachment can be done by any of a variety of chemical means known in the art. In this case, the first member of an affinity pair can optionally contain a functional chemical group that facilitates binding of the first member of the affinity pair to the termini of the first fragments. It will be appreciated that by using this "chemical method", it is possible to attach first members of an affinity pair to both ends of terminal first fragments. Naturally, using the chemical method it is also possible to include the above-described linker (or spacer) nucleotide sequences.

Where a functional chemical group is attached to the first member of the affinity pair, the linker (or spacer) nucleotide sequence is located between the first member of the affinity pair and the chemical functional group.

The second fragments are then exposed to fragmenting restriction enzyme (FRE). The FRE can be any restriction enzyme whose recognition sequence occurs relatively frequently in the genomic DNA of interest. Thus, restriction enzymes having four nucleotide recognition sequence are particularly desirable as FRE. In addition, the FRE should not be sensitive to methylation, i.e., its recognition sequence, at least in eukaryotic DNA should not contain a CpG dinucleotide sequence. Preferably, the FRE recognition sequence should occur at least 10 (e.g., at least: 20; 50; 100; 500; 1,000; 2,000; 5,000; 10,000; 25,000; 50,000; 100,000; 200,000; 500,000; 10⁶; or 10⁷) times more frequently in the genome than does the MMRE recognition sequence. Examples of useful FRE whose recognition sequences consist of four nucleotides include, without limitation, AIuI, Bf al, CviAII, Fatl, HpyCH4V, Msel, NIaIII, or Tsp509I. The recognition sequence for NIaIII is CATG. Exposure of the second fragments to the FRE results in a large number of fragments, the majority of which will have FRE cut sequences at both of their termini and a relatively few with a FRE cut sequence (5' or 3') at one end and the first member of the affinity pair (corresponding to a MMRE cut sequence) at the other end. The latter fragments are referred to herein as third fragments. The third fragments are then exposed to a solid substrate having bound to it the second member of the affinity pair (e.g., avidin, streptavidin, or a functional fragment of either; see Summary section for examples of other useful second members) corresponding to the first member of the affinity pair in the third fragments. The third fragments bind, via the physical interaction between the first and second members of the affinity pair, to the solid substrate. The solid substrate can be any insoluble substance such as plastic (e.g., plastic microtiter well or p^'etri plate bottoms), metal (e.g., magnetic metallic beads), agarose (e.g., agarose beads), or glass (e.g., glass beads or the bottom of a glass vessel such as a glass beaker, test tube, or flask) to which the third fragments can bind and thus be separated from fragments not containing the first member of the affinity pair. Fragments not bound to the solid substrate are removed from the mixture and the solid substrate is optionally rinsed or washed free of any non-specifϊcally bound material. The third fragments bound to the solid substrate are referred to as bound third fragments.

The terminus of the bound third fragment not bound to the solid substrate (referred to herein as the free terminus) is then conjugated to a releasing restriction enzyme (RRE) (also referred to herein sometimes as a tagging enzyme) recognition sequence. This can be achieved by, for example, ligating to the free termini (containing a FRE 5' or 3' cut sequence) releasing moieties containing the FRE 5' or 3 cut sequence and, 5' of the cut sequence, the RRE recognition sequence. Restriction enzymes useful as RRE are those that cut DNA at specific distances (depending on the particular type Hs restriction enzyme) from the recognition sequence, e.g., without limitation, the type Hs and type II . An example of a useful RRE is Mmel that has the following non- palindromic recognition sequence: 5'-TCCPuAC, 3'-AGGPyTG (Pu, purine; Py, pyrimidine) and cuts DNA after the twentieth nucleotide downstream of the TCCPuAc sequence [Boyd et al. (1986) Nucleic Acids Res. 14(13): 5255-5274]. Other useful type Hs restriction enzymes include, without limitation, Bsnfl, Fokl, and AIwI, and useful type IIB restriction enzymes include, without limitation, BsaXI, CspCI, AIoI, PpH, and others listed in Tengs et al. [(2004) Nucleic Acids Research 32(15):e21(pagesl-9)], the disclosure of which is incorporated herein by reference in its entirety.

Releasing moieties can optionally contain, immediately 5' of the RRE recognition sequence, additional nucleotides as an extending sequence. The extending sequence can be of any convenient length, e.g., one to 100 bp, three to 80 bp, five to 70 bp, seven to 60 bp, nine to 50, or 10 to 40 bp. The extending sequence can be, for example, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 26, 37, 38, or 40 bp long.

Conjugating the RRE recognition sequence to the free termini of the bound third fragments results in bound fourth fragments that (a) have RRE recognition sequences at their free termini, and (b) are bound by the first and second members of the affinity pair to the solid substrate. The bound fourth fragments are then exposed to the RRE which cuts the bound fourth fragments at a position that is characteristic of the relevant RRE. In the case of the MmelRRE, the bound fourth fragment is cut on the downstream side of the twentieth nucleotide after the terminal C residue of the TCCPuAC recognition sequence. The exposure results in the release from the solid substrates of a library of fifth fragments. Each of the fifth fragments contains the RRE recognition sequence (and extending sequence if used) and a plurality of bp of the test genomic DNA, including the FRE recognition sequence closest to an unmethylated MMRE recognition sequence. The absolute number of these bp of the test genomic DNA in the fifth fragments will vary from one RRE to another and is, in the case of Mmel, 20 nucleotides. The sequence of genomic DNA in the fifth fragment (but without the FRE recognition sequence) is referred to herein as a MSDK tag. Since the Mmel and NIaIII recognition sequences overlap by one nucleotide, the tags generated using Mmel as the RRE and NIaIII as the FRE are 17 nucleotides long. The greater the number of bp between the RRE recognition sequence and the cutting site of the RRE, the longer the MSDK tags will be. The longer the MSDK tags are, the lower the chances of redundancy due to a plurality of occurrences of the tag sequence in the genome of interest will be. In addition, it will be appreciated that the number of bp between FRE recognition sequences and corresponding MMRE recognition sequences in the genomic DNA of interest will optimally be greater than the number of bp between the RRE recognition sequence and the RRE cut site. However problems arising due to this criterion not being met can be obviated by using the binding moiety method of attaching a first member of an affinity pair to first fragment termini and including in the binding moiety a linker (or spacer) nucleotide sequence of appropriate length (see above); the shorter the distance between the any given FRE recognition sequence and a corresponding MMRE recognition sequence in a genome being analyzed, the longer the linker (or spacer) nucleotide sequence would need to be.

Methods of using α MSDK tag library MSDK libraries generated as described above can be used for a variety of ■ purposes. The first step in most of such methods would be to at least identify the nucleotide sequences of as many MSDK tags obtained in making a library as possible. There are many ways in which this could be done which will be apparent to those skilled in the art.

For example, array technology or the MPSS (massively parallel signature sequencing) method could be exploited for this purpose. Alternatively, the MSDK tag-containing fifth fragments (see above) can be cloned into sequencing vectors (e.g., plasmids) and sequenced using standard sequencing techniques, preferably automated sequencing techniques.

The inventors have used a technique for identifying MSDK tag sequences (see Example 1 below) adapted from the Sequential Analysis of Gene Expression (SAGE) technique [Porter et al. (2001) Cancer Res. 61:5697-5702; Krop et al. (2001) Proc. Natl.

Acad. Sci. U.S.A 98:9796-9801; LaI et al. (1999) Cancer Res. 59:5403-5407; and Boon . et al. (2002) Proc. Natl. Acad. Sci. U.S.A. 99:11287-11292]. This adapted technique involves: (a) adding a DNA ligase enzyme to a library of fifth fragments and thereby ligating pairs of fifth fragments having cohesive RRE-derived ends together to form fifth fragment dimers (also referred to herein as "ditags");

(b) increasing the numbers of individual ditags by PCR using primers whose sequences correspond to nucleotide sequences in extender sequences derived from a releasing moiety (see above);

(c) digesting the PCR-amplifϊed ditags with the FRE used to generate the MSDK library and thereby generating digested ditags lacking the RRE site and extender sequences (if used);

(d) concatamerizing (polymerizing) the ditags using a ligase enzyme (e.g., T4 ligase) to create ditag multimers;

(e) cloning the ditag multimers into sequencing vectors and sequencing the inserts (e.g., by automatic sequencing methods); and

(f) deducing from the ditag multimer sequences the sequences of individual MSDK tags. One of skill in the art will naturally know of ways to modify and adapt the above tag identification procedure to his or her particular requirements. For example, one or more of the steps (e.g., step (b), the ditag amplification step or step (c), the step that removes the RRE recognition site and any extender sequence used) could be omitted.

Having obtained the sequences of some or all of the MSDK tags, there are a number of analyses that could be pursued.

Enumeration of MSDK tags

The numbers of each tag, or a subgroup of tags, in a MSDK library can be computed. Then, for example, optionally having normalized the number of each to the total number of cloned tag sequences obtained, the resulting MSDK profile (consisting of a list of MSDK tags and the abundance (number) of each MSDK tag) can be compared to corresponding MSDK profiles obtained with other cells of interest. In computing the total numbers of individual MSDK tags, where ditags have been amplified by PCR (step (b) above), ditag replicates are deleted from the analysis. Since the chance of any one ditag combination occurring more than once as a result of step (a) above would be extremely low, replicate ditags would likely be due to the PCR amplification procedure. Ways to estimate the numbers of individual tag sequences include the same methods described above for identifying the tag sequences.

The relative abundance (number) of a given MSDK tag obtained gives an indication of the relative frequency at which the nearest MMRE recognition sequence to the FRE recognition sequence associated with the given tag is unmethylated. The higher the number of the MSDK tag obtained, the more frequently that MMRE recognition sequence is unmethylated. Because, by the nature of the method, any given MMRE recognition sequence is correlated with a MSDK tag associated with the nearest FRE recognition sequence upstream of it and with the nearest FRE recognition sequence downstream of it, if any two MMRE recognition sites occur without an appropriate FRE recognition site between them, it will always be possible to discriminate the methylation status (methylated or not methylated) of both the MMRE recognition sites. On the other hand if three MMRE recognition sites occur without an FRE recognition sequence between the first and third, it might not be possible to discriminate the methylation status of the middle MMRE recognition sequence. However, the chances of this occurring can be reduced to essentially zero by chosing a FRE that has a recognition sequence occurring in the genomic DNA of interest much more frequently than the selected MMRE. Indeed prior to the analysis, since generally the sequence of the genome of interest is known, this potential resolution-impairing eventuality can be tested for in advance and overcome by examining the genomic nucleotide sequences and, if necessary, an alternative MMRE-FRE combination can be selected or a plurality of analyses can be performed using a number of different MMRE-FRE combinations.

MSDK tag profiles composed of all the tag sequences obtained in an MSDK analysis, and preferably (but not necessarily) the relative numbers of all the MSDK tags, can be compared to corresponding profiles obtained with other cell types. Corresponding profiles will of course be those generated using the same MMRE, FRE, and RRE and in at least an overlapping part, if not an identical portion, of the relevant genome. Such comparisons can be used, for example, to identify a test cell of interest. For example, a test cell could be a cell of type x, type y, or type z. The MSDK profile obtained with the test cell can be compared to control corresponding MSDK profiles obtained from control cells of type x, type y, and type z. The test cell will likely be of the same type, or at least most closely related, to the control cell (type x, y, or z) whose MSDK profile the test cell's profile most closely resembles. Alternatively, the MSDK profile of a test cell can be compared to that of a single control cell and, if the test cell's profile is significantly different from that of the control cell's profile, it is likely to be of a different type than the control cell type. Statistical methods for doing the above-described analyses are known to those skilled in the art.

The number of MSDK tag species in any given MSDK tag profile varies greatly depending on how many are available and their relative discriminatory power. Indeed, where a particular MSDK tag can discriminate specifically between two cell types of interest, the MSDK tag profile can contain it alone. Thus MSDK tag profiles can contain as few as one MSDK tag. However, they will generally contain a plurality of different MSDK tags , e.g., at least: 2; 3; 4; 5; 6; 7; 8; 9, 10; 12; 15; 20; 25; 30; 35; 40; 50; 60; 75; 85; 100; 120; 140; 160; 180; 200; 250; 300; 350; 400; 450; 500; 600; 700; 800; 900; a 1,000; 2,000; 5,000; 10,000; or even more tag species. The range of "cell types" that can be compared in the above analyses is of course enormous. Thus, for example, the MSDK profile of a test bacterium can be compared to control MSDK profiles of bacteria of: various species of the same genus as the test bacterium (if its genus is known but its species is to be defined); various strains of the same species as the test bacterium (if its species is known but its strain is to be defined) or even various isolates of the same strain as the test bacterium but from, for example, various ecological niches (if the strain of the test bacterium, but not its ecological origin, is known). The same principle can be applied to any biological cell and to any level of speciation of a biological cell. Similarly the MSDK profiles of eukaryotic (e.g., mammalian) test cells can be compared to corresponding MSDK profiles of control test cells of various tissues, of various stages of development, and of various lineages. In addition, the MSDK profile of a test vertebrate cell can be compared to one or more control MSDK profiles of cells (of, for example, the same tissue as the test cell) that are normal or malignant in order to determine (diagnose) whether the test cell is a malignant cell. Moreover, the MSDK profile of a cancer test cell can be compared to one or more control MSDK profiles of cancers of a variety of tissues in order to define the tissue origin of the test cell. In addition, the MSDK profile of a test cell can be compared to that or those of (a) control test cell(s) that can be identical to, or similar to or even different from, the test cell but has/have been exposed or subjected to any of large number of experimental or natural influences, e.g., drugs, cytokines, growth factors, hormones, or any other pharmaceutical or biological agents, physical influences (e.g., elevated and/or depressed temperature or pressure), or environmental conditions (e.g., drought or monsoon conditions). It will thus be appreciated that the term "cell type" covers a large variety of cells and that (or those) used or defined in any particular analysis will depend on the nature of analysis being performed. Those skilled in the art will be able to select appropriate control cell types for the analyses of interest. Examples of MSDK profiles useful as control test profiles are provided herein.

Thus, for example, the MSDK profile of a test breast cell (e.g., an epithelial cell, a myoepithelial cell, or a fibroblast) from a human subject could be compared to the MSDK profiles of breast epithelial cells, myoepithelial cells, and fibroblast-enriched stromal cells from both control normal and control breast cancer (e.g., DCIS or invasive breast cancer) subjects in order to establish whether the test breast tissue from which the test breast cell was obtained is cancerous breast tissue. Moreover, the MSDK profile of a test cancer cell can be compared to those of control breast, prostate, colon, lung, and pancreatic cancer cells as part of an analysis to establish the tissue of the test cancer cell. In addition, the MSDK profile of a cell suspected of being either an epithelial or myoepithelial cell can be compared to those of control normal (and/or cancerous, depending on whether the test cell is normal, cancerous, or not yet established to be normal or cancerous) epithelial and myoepithelial cells in order to establish whether the test cell is an epithelial or myoepithelial cell.

Mapping of MMRE recognition sequences Alternatively, or in addition to enumerating MSDK tags, once the tags obtained in by the MSDK analysis have been identified, the locations in the genome of interest corresponding to the tags (referred to herein as "genomic tag sequences) can be established by comparison of the tag sequences to the nucleotide sequence of the genome (or part of the genome) of interest. This can be done manually but is preferably done by computer. The relevant genomic sequence information can be loaded into the computer from a medium (e.g., a computer diskette, a CD ROM, or a DVD) or it can be downloaded from a publicly available internet database.

One method by which the genomic tag sequences can be identified is by first creating a "virtual" tag library using the following information: (a) the nucleotide sequence of the genome (or part of the genome) of interest; (b) the nucleotide sequence of the MMRE recognition sequence; (c) the nucleotide sequence of the FRE recognition sequence; and (d) the number of nucleotides separating the RRE recognition sequence from the RRE cutting site. Optimally, virtual tag sequences that are not unique (i.e. that could arise in a MSDK library from more than one genetic locus) are deleted from the virtual MSDK library. By comparing the sequences of the tags obtained in the test

MSDK analysis to the virtual tag library, it is possible to determine the genomic location of MSDK tags of interest, e.g., all the tags obtained by the analysis or one or more of such tags.

Once the genomic location of the genomic tag sequences has been obtained, it is a simple matter to identify genes in which, or close to which, the genomic tag sequences are located. This step can be done manually, but can also be done by a computer. Such genes can be the subject of additional analyses, e.g., those described below.

Methods of determining levels of DNA methylation The invention features methods of assessing the level of methylation of genomic regions (e.g., genes or subregions of genes) of interest. The methods can be applied to genomic regions identified by the MSDK analyses described above or selected on any other basis, e.g., the observation of differential expression of a gene in two cell types (e.g., a normal cell and a cancer cell of the same tissue as the normal cell) of interest. The methods are of particular interest in the diagnosis of cancer. In broad terms, it has been claimed that the genomes of cancer cells are hypomethylated relative to corresponding normal cells [Feinberg et al. (1983) Nature 301:89-92]. Moreover, gene hypermethylation is frequently associated with decreased expression of the relevant gene. However, at the individual gene level these generalizations do not apply. Thus, for example, some genes can be hypermethylated in cancer cells in comparison to corresponding normal cells, hypermethylation of some genes is associated with increased expression, and hypomethylation of some genes is associated with decreased expression of the relevant genes. Interestingly, in the examples below, it was observed that hypermethylation of the promoter region of one gene (Cxorfl2) was associated with decreased expression of the gene, while hypermethylation of the exons and/or introns of three other genes (PRDM14, HOXD4, and CDC42EP5 ) was associated with increased expression of the genes.

As used herein, the term "gene" refers to a genomic region starting 10 kb (kilobases) 5' of a transcription initiation site and terminating 2 kb 3' of the polyA signal associated with the coding sequence within the genomic region. Where the polyA signal of another gene is located less than 10 kb 5' of the transcription initiation site of a gene of interest, for the purposes of the instant invention, the gene of interest is considered to start at the first nucleotide immediately after the polyA signal of the other gene. Moreover, where a transcription initiation site of another gene is less than 2 kb 3' prime of the polyA signal of the gene of interest, for the purposes of the instant invention, the gene of interest terminates at the nucleotide immediately before the transcription initiation site of the other gene. From these definitions it will be appreciated that, as used herein, promoter regions and regions 3' of polyA signals of adjacent genes can overlap.

As used herein, the "promoter region" of a gene refers to a genomic region starting 10 kb 5 ' of a transcription initiation site and terminating at the nucleotide immediately 5' of the transcription initiation site. Where a polyA signal of another gene is located less than 10 kb 5' of the transcription initiation site of a gene of interest, for the purposes of the instant invention, the promoter region of the gene of interest starts at the first nucleotide immediately following the polyA signal of the other gene. As used herein, the terms "exons" and "introns" refer to amino acid coding and non-coding, respectively, nucleotide sequences occurring between the transcription initiation site and start of the polyA sequence of a gene.

As used herein, a "CpG island" is a sequence of genomic DNA in which the number of CpG dinucleotide sequences is significantly higher than their average frequency in the relevant genome. Generally, CpG islands are not greater than 2,000 (e.g., not greater than: 1,900; 1,800; 1,700; 1,600; 1,500; 1,400; 1,300; 1,200; 1,100; 1,000; 900; 800; 700; 600; 500; 400; 300; 200; 100; 75; 50; 25; or 15) bp long. They will generally contain not less than one CpG sequence to every 100 (e.g., every: 90; 80; 70; 60; 50; 40; 35; 30; 25; 20; 15; 10; or 5) bp in sequence of DNA. CpG islands can be separated by at least 20 (i.e., at least: 20; 35; 50; 60; 80; 100; 150; 200; 250; 300; 350; or 500) bp of genomic DNA.

In the methods of the invention, the degree of methylation of one or more C residues (in CpG sequences) in a gene of a test cell is determined. This degree of methylation can then be compared to that in one or more (e.g., two, three, four, five, six, seven, eight, nine, ten, 11, 12, 15, 18, 20, 25, 30, 35, 40, 50, 75, 100, 200, or more) control cells.

If the level of methylation in the test cell is altered compared to, for example, that of a control cell, the test cell is likely to be different from the control cell. For example, the test cell can be a cell from any of the vertebrate tissues recited herein, the control cell can be a normal of that tissue, and the gene can be any one that is differentially methylated in cells from cancerous versus normal tissue (e.g., any of the genes listed in Tables 2, 5, 7, 8, 10, 12 and 15). If the degree of methylation of the gene in the test cell is different from that in the normal cell, the test cell is likely to be a cancer cell.

Alternatively, the level of methylation in the test cell can be compared to that in two more (see above) control cells. The cell will be the same as, or most closely related to, the control cell in which the degree of methylation is the same as, or most closely resembles, that of the test cell.

The whole of a gene or parts of a gene (e.g., the promoter region, the transcribed regions, the translated region, exons, introns, and/or CpG islands) can be analyzed. Test and control cells can be the same as those listed above in the section on MSDK. Genes that can analyzed can be any gene differently methylated in two or more cell types of interest. In the methods of the invention any number of genes can be analyzed in order to characterize a test cell of interest. Thus, one, two, three, four, five, six, seven, eight, nine, ten, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 22, 25, 28, 30, 35, 40, 45, 50, 60, 70, 80, 80, 100, 200, 500, or even more genes can be analyzed. The genes can be, for example, any of the DNA sequences (e.g., the genes) listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16. The entire genes or one more subregions of the genes (e.g., all or parts of promoter regions, all or parts of transcribed regions, exons, introns, and regions 3' of polyA signals) can be analyzed

Specific genes of interest include, for example, the LMX- 14, COL5A, LHX3, TCF7L1, PRDM14, ZCCHC14, HOXD4, SLC9A3R1, CDC42EP5, Cxorfl2,

LOC389333, SOX13, SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and HOXAlO genes.

Methylation levels of one or more of these DNA sequences (e.g., genes) can be used to determine, for example, whether a test epithelial cell from breast tissue is a normal or cancerous epithelial cell (e.g., a DCIS (high, intermediate, or low grade) or invasive breast cancer cell). Particularly useful for such determinations are the PRDM14 and ZCCHC 14 genes. For example, with respect to the PRDM 14 gene, a gene segment that is or contains all or part of SEQ ID NO: 1 (Fig. 6A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 8 - 17; 341-392; 371-426; or 391-405 of SEQ ID

NO: 1. Methylation of the PRDM14 can similarly be used to determine whether a test cell from, for example, pancreas, lung, or prostate is a cancer cell or normal cell. In addition, with respect to the ZCCHC 14 gene, a gene segment that is or contains all or part of SEQ ID NO:2 (Fig. 17) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 154-236; 154- 279; 154-293; or 154-299 of SEQ ID NO:2. Hypermethylation of these genes, and particularly hypermethylation of their coding regions, would indicate that the relevant test cells are cancer cells.

In addition, methylation levels of one or more of the above-listed genes can be used to determine, for example, whether a test epithelial cell from colon tissue is a normal or cancerous epithelial cell. Particularly useful for such determinations are the LHX3, TCF7L1, and LMX-IA genes. For example, with respect to the LHX3 gene, a gene segment that is or contains all or part of SEQ ID NO:3 (Fig. 6A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 667-778; 739-788; 918-931; or 885-903 of SEQ ID NO:3. In addition, for example, with respect to the TCF7L1 gene, a gene segment that is or contains all or part of SEQ ID NO:4 (Fig. 8A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 708-737; 761-780; 807-864; or 914-929 of SEQ ID NO:4. Moreover, for example, with respect to the LMX-IA gene, a gene segment that is or contains all or part of SEQ ID NO: 5 (Fig. 7A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 849-878; 898-940; 948-999; or 1,020-1039 of SEQ ID NO:5. Hypermethylation of these genes would indicate that the test cell is a cancerous colon epithelial cell. Furthermore, methylation levels of the above-listed genes can be analyzed to determine, for example, whether breast tissue from which a test myoepithelial is obtained is normal or cancerous breast tissue. Particularly useful for such determinations are the HOXD4, SLC9A3R1 , and CDC42EP5 genes. For example, with respect to the HOXD4 gene, a gene segment that is or contains all or part of SEQ ID NO:6 (Fig. 18A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 185-255; 288-313; 312-362; or 328- 362 of SEQ ID NO:6. In addition, for example, with respect to the SLC9A3R1 gene, a gene segment that is or contains all or part of SEQ ID NO:7 (Fig. 19A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 104-126; 104-247; 104-283; or 246-283 of SEQ ID NO:7. Moreover, for example, with respect to the CDC42EP5 gene, a gene segment that is or contains all or part of SEQ ID NO:8 (Fig. 2 IA) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides: 181-247; 282-328; 336-359; or 336- 390 of SEQ ID NO: 8. Hypermethylation of these genes, and particularly their coding regions, would indicate that the test myoepithelial cell is from cancerous breast tissue. Methylation levels of the above-listed genes can also be analyzed to determine, for example, whether breast tissue from which a test fibroblast is obtained is normal or cancerous breast tissue. Particularly useful for such determinations is the Cxorfl2 gene. For example, with respect to the either of these genes, a gene segment that is or contains all or part of SEQ ID NO: 9 (Fig. 22A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose nucleotide sequences that include nucleotides: 120-134; 159-201; 206-247; or 293-313 of SEQ ID NO:9. Hypermethylation of these genes, and particularly their promoter regions, would indicate that the test fibroblast is from cancerous breast tissue. In addition, methylation levels of the above-listed genes can also be analyzed to determine, for example, whether a test cell is an epithelial cell or a myoepithelial cell. Such assays can be applied to both normal and cancerous cells. Particularly useful for such determinations are the LOC389333 and CDC42EP5 genes. For example, with respect to the LOC389333 gene, a gene segment that is or contains all or part of SEQ ID NO: 10 (Fig. 20A) can be analyzed in order to discriminate these cell types. Of particular interest for this purpose are nucleotide sequences that include nucleotides:306-330; 334- 361; 373-407; or 415-484 of SEQ E) NO:10. With respect to the CDC42EP5 gene, examples of gene segments that can be analyzed include those described above for discriminating whether tissue from which a test myoepithelial was obtained was normal or cancerous. Significantly high levels of methylation of these genes would indicate that the test cell was an epithelial rather than a myoepithelial cell. In addition, methylation levels of the above-listed genes can also be analyzed to determine, for example, whether a test cell is a stem cell, or a differentiated cell derived therefrom, such as an epithelial cell or a myoepithelial cell. Such assays can be applied to both normal and cancerous cells. Particularly useful for such determinations are the SOX13, SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and

HOXAlO genes. For example, with respect to the FOXCl gene, a gene segment that is or contains all or part of SEQ ID NO: 12 (Fig. 27A) can be analyzed in order to discriminate these cell types. In some cases, significantly high levels of methylation of some of these genes would indicate that the test cell was a stem cell rather than a differentiated cell derived therefrom, (e.g., an epithelial or a myoepithelial cell).

Levels of methylation of C residues of interest can be assessed and expressed in quantitative, semi-quantitative, or qualitative fashions. Thus they can, for example, be measured and expressed as discrete values. Alternatively, they can be assessed and expressed using any of a variety of semi-quantitative/qualitative systems known in the art. Thus, they can be expressed as, for example, (a) one or more of "very high", "high", "average", "moderate", "low", and /or "very low"; (b) one or more of "++++", "+++", "++", "+", "+/-", and/or "-"; (c) methylated or not methylated (i.e., in a digital fashion); (d) ranges such as "0% -10%", "l l%-20%", 21% - 30%", "31%-40%, etc. (or any convenient range intervals); (e) graphically, e.g., in pie charts. Methods of measuring the degree of methylation of C residues in the CpG sequences are known in the art. Such methodologies include sequencing of sodium bisulfite-treated DNA and methylation-specifϊc PCR and are described in the Examples below.

Standardizing methylation assays to discriminate between cell types of interest involves experimentation entirely familiar and routine to those in the art. For example, the methylation status of gene Q in a sample cancer cells of interest obtained from a one or more patients and in corresponding normal cells from normal individuals or from the same patients can be assessed. From such experimentation it will be possible to establish a range of "cancer levels" of methylation and a range of "normal levels" of methylation of gene Q. Alternatively, the methylation status of gene Q in cancer cells of each patient can be compared to the methylation status of gene Q in normal cells (corresponding to the cancer cells) obtained from the same patient. In such assays, it is possible that methylation of as few as one cytosine residue could discriminate between cancer and non-cancer cells.

Other methods for quantitating methylation of DNA are known in the art. Such methods are based on: (a) the inability of methylation-sensitive restriction enzymes to cleave sequences that contain one or more methylated CpG sites [Issa et al. (1994) Nat.

Genet. 7:536-540; Singer-Sam et al. (1990) MoI. Cell. Biol. 10:4987-4989; Razin et al.

(1991) Microbiol. Rev. 55:451-458; Stoger et al. (1993) Cell 73:61-71]; and (b) the ability of bisulfite to convert cytosine to uracil and the lack of this ability of bisulfite on methylated cytosine [Frommer et al. (1992) Proc. Natl. Acad. Sd. USA 89:1827-1831;

Myόhanen et al. (1994) DNA Sequence 5:1-8; Herman et al. (1996) Proc. Natl. Acad. Sd.

USA 93:9821-9826; Gonzalgo et al. (1997) Nucleic Acids Res. 25:2529-2531; Sadri et al.

(1996) Nucleic Acids Res. 24:5058-5059; Xiong et al. (1997) Nudeic Acids Res. 25:2532-

2534].

Gene expression assays

Experiments described in the Examples herein show that in a first cell in which methylation of a gene is altered (increased or decreased) relative to a second cell, expression of the gene in the first cell is also altered relative to the second cell. In addition, previous findings and the data in the Examples indicate that alterations in methylation status, and hence also consequent alterations in expression, of certain genes correlate with phenotypic changes in cells. These findings provide the basis for assays

(e.g., diagnostic assays) to discriminate between two or more cell types.

In the methods of the invention, the level of expression of a gene of a test cell determined. This level of expression can then be compared to that in one or more (e.g., two, three, four, five, six, seven, eight, nine, ten, 11, 12, 15, 18, 20, 25, 30, 35, 40, 50, 75,

100, 200, or more) control cells.

If the level of expression in the test cell is altered compared to, for example, that of a control cell, the test cell is likely to be different from the control cell. For example, the test cell can be a cell from any of the vertebrate tissues recited herein, the control cell can be a normal cell of that tissue, and the gene can be one shown to be differentially methylated in cells from cancerous and normal tissue (e.g., any of the genes listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16). If the level of expression of the gene in the test cell is different from that in the normal cell, the test cell is likely to be a cancer cell.

Alternatively, the level of expression in the test cell can be compared to that in two more (see above) control cells. The cell will be the same as, or most closely related to, the control cell in which the level of expression is the same as, or most closely resembles that of the test cell.

Test and control cells can be any of those listed above in the section on MSDK. Genes whose level of expression can be determined can be any gene differently methylated in two more cell types of interest. They can be, for example, any of the genes listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16.

Specific genes of interest include the LMX- 14, COL5A, LHX3, TCF7L1, PRDM14, ZCCHC14, HOXD4, SOX13, SLC9A3R1, CDC42EP5, Cxorfl2, and LOC389333 genes. Expression levels of one or more of these genes can be analyzed to determine, for example, whether a test epithelial cell from breast tissue is a normal or cancerous epithelial cell (e.g., a DCIS (high, intermediate, or low grade) or invasive breast cancer cell). Particularly useful for such determinations are the PRDM14 and ZCCHC14 genes. Moreover, expression of the PRDM14 can be used to test whether a test cell from prostate, pancreas, or lung tissue is a cancer cell. Thus, for example, enhanced expression of the PRDM 14 gene, or altered expression of the ZCCHC 14 gene, in the test breast epithelial cell compared to a control normal breast epithelial cell would be an indication that the test epithelial cell is a cancer cell.

In addition, expression levels of one or more of the above-listed genes can be analyzed to determine, for example, whether a test epithelial cell from colon tissue is a normal or cancerous epithelial cell. Particularly useful for such determinations are the LHX3, TCF7L1, and LMX-IA genes. Altered expression of these genes in the test colon epithelial cell compared to a control normal control epithelial cell would be an indication that the test colon epithelial cell is a cancer cell. Expression levels of one or more of the above-listed genes in a test myoepithelial cell can be analyzed to determine, for example, whether breast tissue from which the test myoepithelial was obtained is normal or cancerous breast tissue. Particularly useful for such determinations are the HOXD4, SLC9A3R1, and CDC42EP5 genes. Enhanced expression of, for example, the HOXD4 and CSD42EP5 genes, or altered expression of the SLC9A3R1 gene, in the test myoepithelial cell compared to a control myoepithelial from control normal breast tissue, would indicate that the test breast tissue is cancerous breast tissue.

Expression levels of one or more of the above-listed genes in a test fibroblast can also be analyzed to determine, for example, whether breast tissue from which the test fibroblast was obtained is normal or cancerous breast tissue. Particularly useful for such determinations is the Cxorfl2 gene. Expression, for example, of this gene at the same or a greater level than in a control fibroblast from control normal breast tissue would indicate that the breast tissue is not cancerous breast tissue.

In addition, expression levels of one or more of the above-listed genes can also be analyzed determine, for example, whether a test cell is an epithelial cell or a myoepithelial cell. Such assays can be applied to both normal and cancerous cells.

Particularly useful for such determinations are the LOC3.89333 and CDC42EP5 genes. Expression of these genes in the test cell at level that is the same as or similar to that of a control myoepithelial cell would be an indication that the test cell is a myoepithelial cell. On the other hand, expression of the genes in the test cell at level that is the same as or similar to that of a control epithelial cell would be an indication that the test cell is an epithelial cell.

Levels of expression of genes of interest can be assessed and expressed in quantitative, semi-quantitative, or qualitative fashions. Thus they can, for example, be measured and expressed as discrete values. Alternatively, they can be assessed and expressed using any of a variety of semi-quantitative/qualitative systems known in the art. Thus, they can be expressed as, for example, (a) one or more of "very high", "high", "average", "moderate", "low", and /or "very low"; (b) one or more of "++++", "+++", "++", "+", "+/-", and/or "-"; (c) expressed or not expressed (i.e., in a digital fashion): (d) ranges such as "0% -10%", "ll%-20%", 21% - 30%", "31%-40%, etc. (or any convenient range intervals); or (e) graphically, e.g., in pie charts.. In the description below, a "gene X" represents any of the genes listed in Tables 2, 5, 7, 8, 10, and 12; mRNA transcribed from gene X is referred to as "mRNA X"; protein encoded by gene X is referred to as "protein X"; and cDNA produced from mRNA X is referred to as "cDNA X". It is understood that, unless otherwise stated, descriptions containing these terms are applicable to any of the genes listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16, mRNAs transcribed from such genes, proteins encoded by such genes, or cDNAs produced from the mRNAs.

In the assays of the invention either: (1) the presence of protein X or mRNA X in cells is tested for or their levels in cells are assessed; or (2) the level of protein X is assessed in a liquid sample such as a body fluid (e.g., urine, saliva, semen, blood, or serum or plasma derived from blood); a lavage such as a breast duct lavage, lung lavage, a gastric lavage, a rectal or colonic lavage, or a vaginal lavage; an aspirate such as a nipple aspirate; or a fluid such as a supernatant from a cell culture. In order to test for the presence, or measure the level, of mRNA X in cells, the cells can be lysed and total RNA can be purified or semi-purified from lysates by any of a variety of methods known in the art. Methods of detecting or measuring levels of particular mRNA transcripts are also familiar to those in the art. Such assays include, without limitation, hybridization assays using detectably labeled mRNA X-specific DNA or RNA probes and quantitative or semi-quantitative RT-PCR methodologies employing appropriate mRNA X and cDNA X-specific oligonucleotide primers. Additional methods for quantitating mRNA in cell lysates include RNA protection assays and serial analysis of gene expression (SAGE). Alternatively, qualitative, quantitative, or semi-quantitative in situ hybridization assays can be earned out using, for example, tissue sections or unlysed cell suspensions, and detectably (e.g., fluorescently or enzyme) labeled DNA or RNA probes. Methods of detecting or measuring the levels of a protein of interest in cells are known in the art. Many such methods employ antibodies (e.g., polyclonal antibodies or monoclonal antibodies (mAbs)) that bind specifically to the protein. In such assays, the antibody itself or a secondary antibody that binds to it can be detectably labeled. Alternatively, the antibody can be conjugated with biotin, and detectably labeled avidin (a protein that binds to biotin) can be used to detect the presence of the biotinylated antibody. Combinations of these approaches (including "multi-layer" assays) familiar to those in the art can be used to enhance the sensitivity of assays. Some of these assays (e.g., immunohistological methods or fluorescence flow cytometry) can be applied to histological sections or unlysed cell suspensions. The methods described below for detecting protein X in a liquid sample can also be used to detect protein X in cell lysates. Methods of detecting protein X in a liquid sample (see above) basically involve contacting a sample of interest with an antibody that binds to protein X and testing for binding of the antibody to a component of the sample. In such assays the antibody need not be detectably labeled and can be used without a second antibody that binds to protein X. For example, by exploiting the phenomenon of surface plasmon resonance, an antibody specific for protein X bound to an appropriate solid substrate is exposed to the sample. Binding of protein X to the antibody on the solid substrate results in a change in the intensity of surface plasmon resonance that can be detected qualitatively or quantitatively by an appropriate instrument, e.g., a Biacore apparatus (Biacore International AB, Rapsgatan, Sweden). Moreover, assays for detection of protein X in a liquid sample can involve the use, for example, of: (a) a single protein X-specific antibody that is detectably labeled; (b) an unlabeled protein X-specific antibody and a detectably labeled secondary antibody; or (c) a biotinylated protein X-specific antibody and detectably labeled avidin. In addition, as described above for detection of proteins in cells, combinations of these approaches (including "multi-layer" assays) familiar to those in the art can be used to enhance the sensitivity of assays. In these assays, the sample or an (aliquot of the sample) suspected of containing protein X can be immobilized on a solid substrate such as a nylon or nitrocellulose membrane by, for example, "spotting" an aliquot of the liquid sample or by blotting of an electrophoretic gel on which the sample or an aliquot of the sample has been subjected to electrophoretic separation. The presence or amount of protein X on the solid substrate is then assayed using any of the above-described forms of the protein X-specific antibody and, where required, appropriate detectably labeled secondary antibodies or avidin.

The invention also features "sandwich" assays. In these sandwich assays, instead of immobilizing samples on solid substrates by the methods described above, any protein X that may be present in a sample can be immobilized on the solid substrate by, prior to exposing the solid substrate to the sample, conjugating a second ("capture") protein X- specific antibody (polyclonal or mAb) to the solid substrate by any of a variety of methods known in the art. hi exposing the sample to the solid substrate with the second protein X-specific antibody bound to it, any protein X in the sample (or sample aliquot) will bind to the second protein X-specific antibody on the solid substrate. The presence or amount of protein X bound to the conjugated second protein X-specific antibody is then assayed using a "detection" protein X-specific antibody by methods essentially the same as those described above using a single protein X-specific antibody. It is understood that in these sandwich assays, the capture antibody should not bind to the same epitope (or range of epitopes in the case of a polyclonal antibody) as the detection antibody. Thus, if a mAb is used as a capture antibody, the detection antibody can be either: (a) another mAb that binds to an epitope that is either completely physically separated from or only partially overlaps with the epitope to which the capture mAb binds; or (b) a polyclonal antibody that binds to epitopes other than or in addition to that to which the capture mAb binds. On the other hand, if a polyclonal antibody is used as a capture antibody, the detection antibody can be either (a) a mAb that binds to an epitope to that is either completely physically separated from or partially overlaps with any of the epitopes to which the capture polyclonal antibody binds; or (b) a polyclonal antibody that binds to epitopes other than or in addition to that to which the capture polyclonal antibody binds. Assays which involve the use of a capture and detection antibody include sandwich ELISA assays, sandwich Western blotting assays, and sandwich immunornagnetic detection assays.

Suitable solid substrates to which the capture antibody can be bound include, without limitation, the plastic bottoms and sides of wells of microtiter plates, membranes such as nylon or nitrocellulose membranes, polymeric (e.g., without limitation, agarose, cellulose, or polyacrylamide) beads or particles. It is noted that protein X-specific antibodies bound to such beads or particles can also be used for immunoaffmity purification of protein X.

Methods of detecting or for quantifying a detectable label depend on the nature of the label and are known in the art. Appropriate labels include, without limitation, radionuclides (e.g., ¹²⁵1, ¹³¹1, ³⁵S, ³H, ³²P, ³³P, or ¹⁴C), fluorescent moieties (e.g., fluorescein, rhodamine, or phycoerythrin), luminescent moieties (e.g., Qdot™ nanoparticles supplied by the Quantum Dot Corporation, Palo Alto, CA), compounds that absorb light of a defined wavelength, or enzymes (e.g., alkaline phosphatase or horseradish peroxidase). The products of reactions catalyzed by appropriate enzymes can be, without limitation, fluorescent, luminescent, or radioactive or they may absorb visible or ultraviolet light. Examples of detectors include, without limitation, x-ray film, radioactivity counters, scintillation counters, spectrophotometers, colorimeters, fluorometers, luminometers, and densitometers.

In assays, for example, to diagnose breast cancer, the level of protein X in, for example, serum (or a breast cell) from a patient suspected of having, or at risk of having, breast cancer is compared to the level of protein X in sera (or breast cells) from a control subject (e.g., a subject not having breast cancer) or the mean level of protein X in sera (or breast cells) from a control group of subjects ( e.g., subjects not having breast cancer). A significantly higher level, or lower level (depending on whether the gene of interest is expressed at higher or lower level in breast cancer or associated stromal cells), of protein X in the serum (or breast cells) of the patient relative to the mean level in sera (or breast cells) of the control group would indicate that the patient has breast cancer.

Alternatively, if a sample of the subject's serum (or breast cells) that was obtained at a prior date at which the patient clearly did not have breast cancer is available, the level of protein in the test serum (or breast cell) sample can be compared to the level in the prior obtained sample. A higher level, or lower level (depending on whether the gene of interest is expressed at higher or lower level in breast cancer or associated stromal cells) in the test serum (or breast cell) sample would be an indication that the patient has breast cancer. Moreover, a test expression profile of a gene in a test cell (or tissue) can be compared to control expression profiles of control cells (or tissues) previously established to be of defined category (e.g., DCIS grade, breast cancer stage, or state of differentiation). The category of the the test cell (or tissue) will be that of the control cell (or tissue) whose expression profile the test cell's (or tissue's) expression profile most closely resembles. These expression profile comparison assays can be used to compare any of the normal breast tissue with any stage and/or grade of breast cancer recited herein and/or to compare between breast cancer grades and stages. The genes analyzed can be any of those listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16 and the number of genes analyzed can be any number, i.e., one or more. Generally, at least two (e.g., at least: two; three; four; five; six; seven; eight; nine; ten; 11; 12; 13; 14; 15; 17; 18; 20; 23; 25; 30; 35; 40; 45; 50; 60; 70; 80; 90; 100; 120; 150; 200; 250; 300; 350; 400; 450; 500; or more) genes will be analyzed. It is understood that the genes analyzed will include at least one of those listed herein but can also include others not listed herein.

One of skill in the art will appreciate from this description how similar "test level" versus "control level" comparisons can be made between other test and control samples described herein.

It is noted that the patients and control subjects referred to above need not be human patients. They can be for example, non-human primates (e.g., monkeys), horses, sheep, cattle, goats, pigs, dogs, guinea pigs, hamsters, rats, rabbits or mice.

Arrays and kits and uses thereof

The invention features an array that includes a substrate having a plurality of addresses. At least one address of the plurality includes a capture probe that binds specifically to any of the MSDK tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16, a nucleic acid X (e.g., a DNA sequence (Ascl site) defined by the location of the MSDK tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16), or a protein X. The array can have a density of at least, or less than, 10, 20 50, 100, 200, 500, 700, 1,000, 2,000, 5,000 or 10,000 or more addresses/cm², and ranges between. In a preferred embodiment, the plurality of addresses includes at least 10, 100, 500, 1,000, 5,000, 10,000, 50,000 addresses. In a preferred embodiment, the plurality of addresses includes equal to or less than 10, 100, 500, 1,000, 5,000, 10,000, or 50,000 addresses. The substrate can be a two- dimensional substrate such as a glass slide, a wafer (e.g., silica or plastic), a mass spectroscopy plate, or a three-dimensional substrate such as a gel pad. Addresses in addition to address of the plurality can be disposed on the array.

An array can be generated by any of a variety of methods. Appropriate methods include, e.g., photolithographic methods (see, e.g., U.S. Patent Nos. 5,143,854;

5,510,270; and 5,527,681), mechanical methods (e.g., directed-flow methods as described in U.S. Patent No. 5,384,261), pin-based methods (e.g., as described in U.S. Pat. No. 5,288,514), and bead-based techniques (e.g., as described in PCT US/93/04145).

In one embodiment, at least one address of the plurality includes a nucleic acid capture probe that hybridizes specifically to any of the MSDK tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16, e.g., the sense or anti-sense (complement) strand of the tag sequences. Each address of the subset can include a capture probe that hybridizes to a different region of the MSDK tag. Such an array can be useful, for example, for detecting the presence and, optionally, assessing the relative numbers of one or more of the MSDK tags (or the complements thereof) listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16 in a sample, e.g., a MSDK tag library.

In another embodiment, at least one address of the plurality includes a nucleic acid capture probe that hybridizes specifically to a nucleic acid X, e.g., the sense or anti- sense strand. Nucleic acids of interest include, without limitation, all or part of any of the genes identified by the tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16, all or part of mRNAs transcribed from such genes, or all or part of cDNA produced from such mRNA. Each address of the subset can include a capture probe that hybridizes to a different region of a nucleic acid. Each address of the subset is unique, overlapping, and complementary to a different variant of gene X (e.g., an allelic variant, or all possible hypothetical variants). The array can be used, for example, to sequence gene X, mRNA X, or cDNA X by hybridization (see, e.g., U.S. Patent No. 5,695,940) or assess levels of expression of gene X.

In another embodiment, at least one address of the plurality includes a polypeptide capture probe that binds specifically to protein X or fragment thereof. The polypeptide can be a naturally-occurring interaction partner of protein X, e.g., a ligand for protein X where protein X if a receptor or a receptor for protein X where protein X is ligand. Preferably, the polypeptide is an antibody, e.g., an antibody specific for protein X, such as a polyclonal antibody, a monoclonal antibody, or a single-chain antibody.

Antibodies can be polyclonal or monoclonal antibodies; methods for producing both types of antibody are known in the art. The antibodies can be of any class (e.g., IgM, IgG, IgA, IgD, or IgE) and be generated in any of the species recited herein. They are preferably IgG antibodies. Recombinant antibodies, such as chimeric and humanized monoclonal antibodies comprising both human and non-human portions, can also be used in the methods of the invention. Such chimeric and humanized monoclonal antibodies can be produced by recombinant DNA techniques known in the art, for example, using methods described in Robinson et al, International Patent Publication PCT/US86/02269; Aldra et al., European Patent Application 184, 187; Taniguchi, European Patent

Application 171,496; Morrison et al., European Patent Application 173,494; Neuberger et al., PCT Application WO 86/01533; Cabilly et al., U.S. Patent No. 4,816,567; Cabilly et al., European Patent Application 125,023; Better et al. (1988) Science 240, 1041-43; Liu et al. (1987) J. Immunol. 139, 3521-26; Sun et al. (1987) PNAS 84, 214-18; Nishimura et al. (1987) Cane. Res. 47, 999-1005; Wood et al. (1985) Nature 314, 446-49; Shaw et al. (1988) J. Natl. Cancer Inst. 80, 1553-59; Morrison, (1985) Science 229, 1202-07; Oi et al. (1986) BioTechniques 4, 214; Winter, U.S. Patent No. 5,225,539; Jones et al. (1986) Nature 321, 552-25; Veroeyan et al. (1988) Science 239, 1534; and Beidler et al. (1988) J. Immunol. 141, 4053-60. Also useful for the arrays of the invention are antibody fragments and derivatives that contain at least the functional portion of the antigen-binding domain of an antibody. Antibody fragments that contain the binding domain of the molecule can be generated by known techniques. Such fragments include, but are not limited to: F(ab')₂ fragments that can be produced by pepsin digestion of antibody molecules; Fab fragments that can be generated by reducing the disulfide bridges of F(ab')₂ fragments; and Fab fragments that can be generated by treating antibody molecules with papain and a reducing agent. See, e.g., National Institutes of Health, 1 Current Protocols In Immunology, Coligan et al., ed. 2.8, 2.10 (Wiley Interscience, 1991). Antibody fragments also include Fv fragments, i.e., antibody products in which there are few or no constant region amino acid residues. A single chain Fv fragment (scFv) is a single polypeptide chain that includes both the heavy and light chain variable regions of the antibody from which the scFv is derived. Such fragments can be produced, for example, as described in U.S. Patent No. 4,642,334, which is incorporated herein by reference in its entirety. For a human subject, the antibody can be a "humanized" version of a monoclonal antibody originally generated in a different species. In another aspect, the invention features a method of analyzing the expression of gene X. The method includes providing an array as described above; contacting the array with a sample and detecting binding of a nucleic acid X or protein X to the array. In one embodiment, the array is a nucleic acid array. Optionally the method further includes amplifying nucleic acid from the sample prior or during contact with the array.

In another embodiment, the array can be used to assay gene expression in a tissue to ascertain tissue specificity of genes in the array, particularly the expression of gene X. If a sufficient number of diverse samples is analyzed, clustering (e.g., hierarchical clustering, k-means clustering, Bayesian clustering and the like) can be used to identify other genes which are co-regulated with gene X. For example, the array can be used for the quantitation of the expression of multiple genes. Thus, not only tissue specificity, but also the level of expression of a battery of genes in the tissue is ascertained. Quantitative data can be used to group (e.g., cluster) genes on the basis of their tissue expression per se and level of expression in that tissue. For example, array analysis of gene expression can be used to assess gene X expression in one or more cell types (see above).

In another embodiment, the array can be used to monitor expression of one or more genes in the array with respect to time. For example, samples obtained from different time points can be probed with the array. Such analysis can identify and/or characterize the development of a gene X-associated disease or disorder (e.g., breast cancer such as invasive breast cancer); and processes, such as a cellular transformation associated with a gene X-associated disease or disorder. The method can also evaluate the treatment and/or progression of a gene X-associated disease or disorder

The array is also useful for ascertaining differential expression patterns of one or more genes in normal and abnormal (e.g., malignant) cells. This provides a battery of genes (e.g., including gene X) that could serve as a molecular target for diagnosis or therapeutic intervention.

In another aspect, the invention features a method of analyzing a plurality of probes. The method is useful, e.g., for analyzing gene expression. The method includes: providing a first two dimensional array having a plurality of addresses, each address (of the plurality) being positionally distinguishable from each other address (of the plurality) having a unique capture probe, e.g., wherein the capture probes are from a cell or subject which express gene X or from a cell or subject in which a gene X-mediated response has been elicited, e.g., by contact of the cell with nucleic acid X or protein X, or administration to the cell or subject of a nucleic acid X or protein X; providing a second two dimensional array having a plurality of addresses, each address of the plurality being positionally distinguishable from each other address of the plurality, and each address of the plurality having a unique capture probe, e.g., wherein the capture probes are from a cell or subject which does not express gene X (or does not express as highly as in the case of the cell or subject described above for the first array) or from a cell or subject which in which a gene X-mediated response has not been elicited (or has been elicited to a lesser extent than in the first sample); contacting the first and second arrays with one or more inquiry probes (which are preferably other than a nucleic acid X, protein X, or antibody specific for protein X), and thereby evaluating the plurality of capture probes. Binding, e.g., in the case of a nucleic acid, hybridization with a capture probe at an address of the plurality, is detected, e.g., by signal generated from a label attached to the nucleic acid, polypeptide, or antibody.

The invention also features a method of analyzing a plurality of probes or a sample. The method is useful, e.g., for analyzing gene expression. The method includes: providing a first two dimensional array having a plurality of addresses, each address of the plurality being positionally distinguishable from each other address of the plurality having a unique capture probe, contacting the array with a first sample from a cell or subject which express or mis-express gene X or from a cell or subject in which a gene X- mediated response has been elicited, e.g., by contact of the cell with nucleic acid X or protein X, or administration to the cell or subject of nucleic acid X or protein X; providing a second two dimensional array having a plurality of addresses, each address of the plurality being positionally distinguishable from each other address of the plurality, and each address of the plurality having a unique capture probe, and contacting the array with a second sample from a cell or subject which does not express gene X (or does not express as highly as in the case of the as in the case of the cell or subject described for the first array) or from a cell or subject which in which a gene X-mediated response has not been elicited (or has been elicited to a lesser extent than in the first sample); and comparing the binding of the first sample with the binding of the second sample. Binding, e.g., in the case of a nucleic acid, hybridization with a capture probe at an address of the plurality, is detected, e.g., by a signal generated from a label attached to the nucleic acid, polypeptide, or antibody. The same array can be used for both samples or different arrays can be used. If different arrays are used the same plurality of addresses with capture probes should be present on both arrays.

All the above listed capture probes useful for arrays can also be provided in the form of a kit or article of manufacture, optionally also containing packaging materials. In such kits or articles of manufacture, the capture probes can be provided as preformed arrays, i.e., attached to appropriate substrates as described above. Alternatively they can be provided in unattached form.

The capture probes can be supplied in unattached form in any number. Moreover, each capture probe in a kit or article of manufacture can be provided in a separate vessel (e.g., bottle, vial, or package), all the capture probes can be combined in the same vessel, or a plurality of pools of capture probes can be provided, with each pool being provided in a separate vessel. In the kit or article of manufacture there can optionally be instructions (e.g., on the packing materials or in a package insert) on how to use the arrays or unattached capture probes, e.g., on how to perform any of the methods described herein.

The following examples are intended to illustrate, not limit, the invention.

EXAMPLES

Example 1. Materials and Methods Tissue specimens and primary cell cultures

Human breast tumor and fresh, frozen, or formalin fixed, paraffin embedded tumor specimens were obtained from the Brigham and Women's Hospital (Boston, MA), Columbia University (New York, NY), University of Cambridge (Cambridge, UK), Duke University (Durham, NC), University Hospital Zagreb (Zagreb, Croatia), the National Disease Research Interchange (Philadelphia, PA), and the Breast Tumor Bank of the University of Liege (Liege, Belgium). All human tissue was collected without patient identifiers using protocols approved by the Institutional Review Boards of the institutions. In the case of matched tissue samples (i.e., normal and tumor tissue samples obtained from the same individuals), the normal tissue corresponding to the tumor was obtained from the ipsilateral breast several centimeters away from the tumor. Fresh tissue samples were immediately processed for immunomagnetic purification and cell subsets were purified as previously described [Allinen et al. (2004) Cancer Cell 6:17-32 and co-pending U.S. Patent Application Serial No. PCT/US2004/08866, the disclosures of which are incorporated herein by reference in its entirety ]. Following the purification procedure, in some cases the purity of each cell population was confirmed by RT-PCR ^λ and primary cultures of the different cell types were initiated. Primary stromal fibroblasts were cultured in DMEM medium supplemented with 10 % iron fortified bovine calf serum (Hyclone, Logan, UT) prior to lysis and DNA and RNA isolation. Human embryonic stem cells were cultured on feeder layers using established protocols (for example, see, REF). DNA and RNA were isolated from the other cell-types without prior culturing.

RNA and genomic DNA isolation, and cDNA synthesis

RNA (total and polyA) isolation was performed using a μMACS™ kit (Miltenyi Biotec, Auburn, CA) from small numbers of cells, while from large tissue samples, primary cultures and cell lines total RNA was isolated using a guanidium/cesium method [Allinen et al. (2004), supra]. Column flow-through fractions (in the μMACS ™ method) and unprecipitated soluble material (guanidium/cesium method) were used for the purification of genomic DNA using SDS/proteinase K digestion followed by phenol- chloroform extraction and isopropanol precipitation. cDNA synthesis was performed using the OMNI-SCRIPT™ kit form Qiagen (Valencia, CA) following the manufacturer's instructions.

Generation and analysis of MSDK (Methylation Specific Digital Karyotyping) libraries MSDK libraries were generated by a modification of the digital karyotping protocol [Wang et al. (2002) Proc. Natl. Acad. Sci USA 16156-16161]. For each sample, 1-5 μg genomic DNA was sequentially digested with the methylation-sensitive enzyme Ascl and the resulting fragments were ligated at their 5' and 3¹ ends to biotinylated linkers (5 ' -biotin-TTTGCAGAGGTTCGTAATCGAGTTGGGTGG-S ' , 5 ' -phos- CGCGCCACCCAACTCGATTACGAACCTCTGC-3'). The biotinylated fragments were then digested with NIaIII as a fragmenting restriction enzyme. Resulting DNA fragments having biotinylated linkers at their termini were immobilized onto streptavidin- conjugated magnetic beads (Dynal, Oslo, Norway).

The remaining steps were essentially the same as those described for LongSAGE with minor modifications [Allinen et al. (2004) supra; Saha et al. (2002) Nat. Biotechnol. 20:508-512]. Briefly, linkers containing the type Hs restriction enzyme Mmel recognition site were ligated to isolated DNA fragments and the bead bound fragments were cut by the Mmel enzyme 21 base pairs away from the restriction enzyme site, resulting in release from the beads into the surrounding solution of tags containing the Mmel recognition site, a linker and 21 base pairs of test genomic DNA. The tags were ligated to form ditags which are formed between single tags containing 5' and 3'

Mmel digestion (cut) sites (depending on whether the relevant fragment bound to a bead was derived by from an NIaIII site 5 ' or 3 ' of an unmethylated Ascl site). The ditags were expanded by PCR, isolated, and ligated to form concatamers, which were cloned into the pZero 1.0 vector (Invitrogen, Carlsbad, CA) and sequenced. 21-bp tags were extracted and duplicate ditags (arising due to the PCR expansion step) were removed using SAGE 2002 software. P values were calculated based on pair- wise comparisons between libraries using a Poisson-based algorithm [Cai et al. (2004) Genome Biol. 5 :R51 ; Allinen et al. (2004) supra] . Raw tag counts were used for comparing the libraries and calculating p values, but subsequently tag numbers were normalized in order to control for uneven total tag numbers/library (average total tag number 28,456 /library).

In order to determine their chromosomal location, tags that appeared only once in each library were filtered out and matched to a virtual Ascl library derived from a human genome sequence. Human genome sequence and mapping information (JuI. 2003, hglό) were downloaded from UCSC Genome Bioinformatics Site. A virtual Ascl tag library was constructed based on the genome sequence as follows: predicted Ascl sites were located in the genomic sequence, the nearest NIaIII sites in both directions to the Ascl sites were identified, and the corresponding virtual MSDK sequence tags were derived. All virtual tags that were not unique in the genome were removed in order to ensure unambiguous mapping of the data. Genes neighboring the Ascl sites were also identified in order to determine the effect of methylation on their expression. .

Alignment of MSDK, SAGE, and CpG islands across the genome

The frequency of Ascl digestion was calculated as percentage of samples (N-EPI- 17, 1-EPI-7, N-MYOEP-4, D-MYOEP-6, N-STR-I7, 1-STR-7, N-STR-I17, 1-STR-17) having raw tag counts of 2 or more at each predicted Ascl site. SAGE counts from corresponding samples (N-EPI- 1 plus N-EPI-2, 1-EPI-7, N-MYOEP- 1 , D-MYOEP-6, D- MYOEP-7, N-STR-I, N-STR-I17, 1-STR-7) were normalized to tags per 200,000. Gene and CpG island position information were downloaded from UCSC Genome Bioinformatics Site (Human genome sequence and mapping information, JuI. 2003, hglβ). Ascl sites were predicted (as mentioned above) from the genome sequence, and Ascl site frequency, SAGE counts, and CpG island positions were drawn together along all chromosomes.

Bisulfite sequencing, quantitative methylation specific PCR (qMSP), and quantitative RT-PCR (qRT-PCR) To determine the location of methylated cytosines, genomic DNA was bisulfite treated, purified, and PCR reactions were performed as previously described [Herman et al. (1996) Proc. Natl. Acad. Sci. USA 93:9821-0826]. PCR products were "blunt- ended", subcloned into pZEROl.O (Invitrogen), and 4-13 independent colonies were sequenced for each PCR product. Based on the above sequence analysis qMSP PCR primers were designed for the amplification of methylated or unmethylated DNA. Quantitative MSP and RT-PCR amplifications were performed as follows. Template (2-5 ng bisulfite treated genomic DNA or 1 μl cDNA) and primers were mixed with 2x SYBR Green master mix (ABI, CA) in a 25 μl volume and the reactions were performed in ABI 7500 real time PCR system (5O⁰C, 20 sec; 95⁰C, 10 min; 95⁰C, 15 sec, 6O⁰C, 1 min (40 cycles); 95⁰C, 15 sec; 6O⁰C, 20 sec; 95⁰C, 15 sec). Triplicates were performed and average Ct values calculated. The Ct (cycle threshold) value is the PCR cycle number at which the reaction reaches a fluorescent intensity above the threshold which is set in the exponential phase of the amplification (based on amplification profile) to allow accurate quantification. In the case of qMSP, methylation of the samples was normalized to methylation independent amplification of the β-actin (ACTB) gene:

%ACTB= 100χ2^{(ctACTB"ctgene)}. For qRT-PCR expression of the samples was normalized to that of the RPL39 (ribosomal protein L39) gene: %RPL39=10x2^(CtRPL39-^ctgene). Normalizations to the expression of the ribosomal protein Ll 9 (RPL 19) and ribosomal protein S 13 (RPS 13) genes were also performed and gave essentially the same results. Due to the very high abundance of ribosomal protein mRNAs, cDNA was diluted tenfold for these PCR reactions relative to that of specific genes. The frequency of methylation of the PRDM 14 gene in normal and tumor samples was calculated by setting a threshold of methylation as the median+2 x standard deviation value of the relative methylation of the normal samples (excluding the one outlier case; see below). Samples above this value (10.66) were defined as methylated.

Example 2. Methylation Specific Digital Karyotyping (MSDK)

The MSDK protocol used in the experiments described below is schematically depicted in Fig. 2. MSDK is a modification of the digital karyotyping (DK) technique recently developed for the analysis of DNA copy number in a quantitative manner on a genome- wide scale [Wang et al. (2002) supra]. DK is based on two concepts: (i) short (e.g., 21 base pair) sequence tags can be derived from specific locations in the human genome; and (ii) these sequence tags can be directly matched to the human genome sequence. The original DK protocol used Sad as a mapping enzyme and NIaIII as a fragmenting enzyme. Using this enzyme combination the tags were obtained from the two (both 5' and 3') NIaIII sites closest to the Sad sites.

In the MSDK method, instead of Sad, a mapping enzyme that is sensitive to DNA methylation was used. Ascl was chosen because its recognition sequence (GGCGCGCC) has two CpG (potential methylation) sites, is preferentially found in CpG islands associated with transcribed genes rather than repetitive elements [Dai et al. (2002) Genome Res. 12: 1591-1598], and it is a rare cutter enzyme (-5,000 predicted sites/human genome) allowing identification of tags that are highly statistically significantly differentially present in the different libraries at reasonable sequencing depths (20,000-50,000 tags/library). Methylation of either or both methylation sites in an Ascl recognition sequence prevents cutting by Ascl. The use of Ascl and NIaIII as mapping and fragmenting enzymes, respectively, with human genomic DNA, respectively, is expected to result in a total of 7,205 virtual tags (defined as possible tags that can be obtained and uniquely matched to the human genome based on the predicted location of Ascl and NIaIII sites). Since Ascl will cut only unmethylated DNA, the presence of a tag in the MSDK library indicates that the corresponding Ascl site is not methylated, while lack of a virtual tag indicates methylation.

To demonstrate the feasibility of the MSDK method for epigenome profiling, MSDK libraries were generated from genomic DNA isolated from the wild-type HCTl 16 human colon cancer cell line (HCT WT) and its derivative in which both the DNMTl and DNMT3b DNA methyltransferase genes have been homozygously deleted (HCT DKO) [Rh.ee et al. (2002) Nature 416, 552-556]. Due to the deletion of these two DNA methyltransferases, methylation of the genomic DNA in the HCT DKO cells is reduced by greater than 95% relative to the HCT WT cells. Thus, MSDK libraries generated from HCT WT and HCT DKO cells were expected to depict dramatic differences in DNA methylation. 21,278 and 24,775 genomic tags were obtained from the WT and DKO cells, respectively. These tags were matched to a virtual Ascl tag library generated as described in Example 1. Unique tags (7, 126 from the WT and 7,964 tags from the DKO cells) were compared and 219 were identified as being statistically significantly (p<0.05) differentially present in the two libraries (Table 1). 137 and 82 of these tags were more abundant in the DKO and WT libraries, respectively. Correlating with the overall hypomethylation of the genome of DKO cells, almost all of the 137 tags were at least 10 fold more abundant in the DKO library, while nearly all 82 tags showed only 2-5 fold difference between the two libraries. Table 1. Chromosomal location and analysis of the frequency of MSDK tags in the HCT116 WT and DKO MSDK libraries.

Chr, Chromosome.

Virtual tags, the number of MSDK tag species predicted for the indicated chromosome.

Observed Tags, the number of different unique tag species observed in both MSDK libraries for the indicated chromosome.

Variety, the number of different unique tag species for the indicated chromosome and MSDK library.

Copies, the abundance (total number) of all the observed unique tags for the indicated chromosome and MSDK library. 5 Tag Variety Ratio, the ratio of the numbers of unique tag species for the indicated chromosome detected in the indicated two libraries.

Tag Copy Ratio, the ratio of the abundances (total numbers) of all the unique tags for the indicated chromosomes detected in the indicated two libraries.

Differential Tag (P< 0.05), the number of unique tag species observed for the indicated chromosome that were present in higher abundance in the one indicated

10 MSDK library than in the other indicated MSDK library (P< 0.050).

Single nucleotide polymorphism (SNP) array analysis of the DNA samples used for the generation of MSDK libraries demonstrated that the two cell lines are indistinguishable using this technique and the observed differences in MSDK tag numbers are unlikely to be due to underlying overt DNA copy number alterations.

Mapping of the tags to the genome revealed that many of the differentially methylated Ascl sites are located in CpG islands and in promoter areas of genes implicated in development and differentiation including numerous homeogenes (Table 2). Consistent with these results, two of these genes, LMX-IA and COL5A, have previously been found to be differentially methylated between HCT 116 WT and DKO cells, and are also frequently methylated in primary colorectal carcinomas and colon cancer cell lines [Paz et al. (2003) Hum. MoI. Genet. 12:2209-2210]. Similarly SCGB3A1/HIN-1, a gene frequently methylated in multiple cancer types [Shigematsu et al. (2005) Int. J. Cancer 113:600-604; Krop et al. (2004) MoI. Cancer Res. 2:489-494; Krop et al. (2001) Proc. Natl. Acad. Sci. USA 98:9796-9801] was identified as one of most highly significantly differently present tags (Table 2).

Table 2.

MSDK tags significantly (p < 0.050) differentially present in HCTl 16 WT and DKO MSDK libraries and genes associates with the MSDK tags.

DKO and WT, raw abundance (total numbers) of indicated MSDK observed in DKO and WT libraries.

Ratio DKO/WT, ratio of normalized abundances (total numbers) of the indicated tag in the DKO and WT libraries (a minus sign indicates that the indicated number is the reciprocal of the DKOAVT ratio).

P value, the significance of the difference in the raw abundances of the relevant MSDK tag between the two libraries.

Chr, chromosome in which MSDK tag sequence is located.

Gene, gene with which the indicated MSDK tag was associated.

Description, description of the product of the associated gene.

The positions of the Ascl site (recognition sequence) identified by the indicated tag relative to the transcription initiation site (tr. Start) of the gene and the distance of the Ascl site

(recognition sequence) from the transcription initiation site are indicated.

In order to further validate the MSDK technique, three highly differentially present tags were selected from the HCT libraries, the corresponding genomic loci (corresponding to the LHX3, LMX-IA, and TCF7L1 genes) were identified, and sequencing of bisulfite treated genomic DNA (the same as that used for the generation of the MSDK libraries) was performed. In all three cases, the relevant Ascl site was completely methylated in the WT and unmethylated in the DKO cells (Figs. 3-5). In addition, almost all other surrounding CpG showed the same methylation/unmethylation pattern. In Figs 6-8 are shown the nucleotide sequences of regions of these three gene segments of which were subjected to the described methylation-detecting sequencing analysis. These results indicated that the MSDK method is suitable for genome- wide analysis of methylation patterns and the identification of differentially methylated sites.

Example 3. Analysis of MSDK libraries from cell populations isolated from normal and cancerous breast tissue

MSDK libraries were generated from epithelial cells, myoepithelial cells, and fibroblast-enriched stroma isolated from normal breast tissue, in situ (DCIS-ductal carcinoma in situ) breast carcinoma tissue, and invasive breast carcinoma tissue. A detailed description of the samples is in Table 3.

Table 3. List of breast tissue samples used for methylation analyses.

The numbers at the ends of the tissue sample names indicate patients from which the tissue samples were obtained.

Age is the age of the particular patient.

LN indicates whether the carcinoma in the relevant patient had spread to one or more lymph nodes.

ER indicates whether the relevant carcinoma cells expressed the estrogen receptor.

PR indicates whether the relevant carcinoma cells expressed the progesterone receptor.

Her2 indicates whether the relevant carcinoma cells expressed Her2/Neu.

Grade is the histologic grade.

Whenever possible, normal and tumor tissue were derived from the same patient in order to control for possible epigenetic variations due to age, and reproductive and disease status. Fibroblast-enriched stroma were the cells remaining after removal of epithelial cells, myoepithelial cells, leukocytes, and endothelial cells and consist of over 80 % fibroblasts. DNA samples were also analyzed with SNP arrays in order to rule out the possibility of overt DNA copy number alterations.

Pair- wise comparisons and statistical analyses of the MSDK libraries revealed that the largest fraction of highly (>10 fold difference) differentially present tags occurred between normal and tumor epithelial cells and the majority of these tags were more abundant in tumor cells (Tables 4 and 5) correlating with the known overall hypomethylation of the cancer genome [Feinberg et al. (1983) Nature 301: 89-92).

Table 4. Chromosomal location and analysis of the frequency of MSDK tags in the I-EPI-7 and N-EIP-I7 MSDK libraries.

The column headings are as indicated for Table 1.

Table 5.

MSDK tags significantly (p < 0.050) differentially present in N-EPI-I7 and I-EPI-7 MSDK libraries and genes associated with the MSDK tags.

The column headings are as in Table 2 except that the MSDK libraries compared are the N-EPI-I7 and I-EPI-7 libraries (see Table 3 for details of the tissues from which these libraries were made).

Although statistically significant differences were observed, a more similar pattern was observed in the comparison of normal and tumor fibroblast-enriched stroma (Tables 6-8).

Table 6. Chromosomal location and analysis of the frequency of MSDK tags in the I-STR-I7 and I-STR-7 MSDK libraries.

The column headings are as indicated for Table 1.

Table 7.

MSDK tags significantly (p < 0.050) differentially present in N-STR-I7 and I-STR-7 MSDK libraries and genes associated with the MSDK tags.

The column headings are as in Table 2 except that the MSDK libraries compared are the N-STR-I7 and I-STR-7 MSDK libraries (see Table 3 for details of the tissues from which these libraries were made).

Table 8.

MSDK tags significantly (p < 0.050) differentially present in N-STR-I17 and I-STR-17 MSDK libraries and genes associated with the MSDK tags.

The column headings are as in Table 2 except that the MSDK libraries compared are the N-STR-117 and I-STR-17 MSDK libraries (see Table 3 for details of the tissues from which the libraries were made).

The comparison of myoepithelial cells isolated from normal breast tissue to those isolated from in situ carcinoma (DCIS) revealed some dramatic differences and indicated relative hypermethylation of the DCIS myoepithelial cells (Tables 9 and 10).

Table 9.

Chromosomal location and analysis of the frequency of MSDK tags in the N-MYOEP-4 and D-MYOEP-6 MSDK libraries.

The column headings are as indicated for Table 1.

Table 10.

MSDK tags significantly differentially (p < 0.050) present in N-MYOEP-4 and D-MYOEP-6 MSDK libraries and genes associated with the MSDK tags.

The column headings are as in Table 2 except that the MSDK libraries are the N-MYOEP-4 and D-MYOEP-6 MSDK libraries (see Table 3 for details of the tissues from which the libraries were made).

Besides identifying epigenetic differences between normal and tumor tissue, cell type-specific differences in methylation patterns were seen by comparing MSDK libraries generated from normal epithelial and normal myoepithelial cells (Tables 11 and 12). Epithelial and myoepithelial cells are thought to originate from a common bi-potential progenitor cell [Bocker et al. (2002) Lab. Invest. 82:737-746]. The methylation differences observed between these two cell types raise the possibility of their different clonal origin or epigenetic reprogramming of the cells during lineage specific differentiation. Indeed, during embryonic development, epigenetic changes are known to occur in a cell lineage specific manner and play a role in differentiation [Kremenskoy et al. (2003) Biochem. Biophys. Res. Commun. 311:884-890].

Table 11. Chromosomal location analysis of the frequency of MSDK tags in the N-MYOEP-4 and N-EPI-I7 MSDK libraries.

The column headings are as indicated for Table 1.

Table 12.

MSDK tags significantly (p < 0.050) differentially present in N-MYOEP4 and N-EPI-I7 MSDK libraries and genes associated with the MSDK tags.

The column headings are as in Table 2 except that the MSDK libraries compared are the N-MYOEP -4 and N-EPI-I7 MSDK libraries (see Table 3 for details of the tissues from which these libraries were made).

In addition to pair-wise comparison of MSDK libraries, genome- wide analyses of methylation and gene expression patterns were performed by combining MSDK and SAGE (Serial Analysis of Gene Expression) data for each breast cell type. The Ascl cutting frequencies were determined and SAGE tag counts were superimposed (details in Example 1). They were then mapped to the human genome together with all predicted CpG islands and Ascl sites. Based on the combined as well as cell-type-specific MSDK and SAGE analysis, it was determined that highly expressed genes are preferentially located in gene dense areas [Caron et al. (2001) Science 291: 1289-1292] and that these areas correlate with the locations of the most frequently cut (thus unmethylated) Ascl sites. Interestingly, while the ratio of the observed and predicted MSDK tags averaged for all cells tested was nearly equal for most chromosomes, chromosomes X and 17 had a lower and a higher observed/expected tag ratio, respectively, in all samples suggesting overall hyper- and hypo-methylation in these specific chromosomes (Tables 1, 2, and 4- 12).

Example 4. Confirmation of MSDK results by sequencing studies

To confirm the MSDK results, several highly differentially methylated genes from each pair-wise comparison were selected and their methylation was analyzed by performing sequence analysis of bisulfite treated genomic DNA from the same sample that was used for MSDK and also from additional samples obtained from independent patients. These genes included PRDM14 and ZCCHC14 (hypermethylated in tumor epithelial cells), HOXD4 and SLC9A3R1 (hypermethylated in DCIS myoepithelial cells) and LOC389333 (more methylated in myoepithelial than in epithelial cells), CDC42EP5 (hypermethylated in DCIS myoepithelial cells and also different between normal epithelial and myoepithelial cells), and Cxorfl2 (hypermethylated in tumor stroma compared to normal) (Figs. 9-15). Interestingly PRDM14 and HOXD4 were also differentially methylated between HCT 116 WT and DKO cells (unmethylated in DKO) suggesting their potential involvement in multiple tumor types or location in a chromosomal area prone to epigenetic modifications. In all these cases bisulfite sequence analysis confirmed the MSDK results although the absolute frequency of methylation was somewhat variable among samples.

In Figs. 16A-22B are shown the nucleotide sequences of the gene regions that were subjected to the above methylation-detecting sequencing analysis.

Example 5. Determination of frequency and consistency of methylation difference by quantitative methylation specific PCR (qMSP) To determine how frequently and consistently methylation differences in these selected genes occur, a quantitative methylation specific PCR (qMSP) assay was developed for some of the genes and their methylation status in a larger set of samples and in multiple cell types was analyzed. This assay depends on the relative ability of two sets of PCR primers targeting segments of DNA that include at least one CpG sequence to anneal to bisulfite treated DNA and cause the amplification of the sequence that the primers span. One set of primers is designed to anneal to the target sequences efficiently and cause the relatively rapid amplification if the target sequences in the DNA are not methylated and the other pair of primers is designed to act similarly if the target sequences in the DNA are methylated.

This analysis not only confirmed the original MSDK data and the bisulfite sequencing results, but also revealed the methylation status of each gene in all three cell types both in normal and tumor tissue (Figs. 23A-E). The frequency of PRDM14 methylation was further analyzed in a panel of normal breast tissue (purified organoids), benign breast tumors (fibroadenomas, fibrocystic dysplasias, and papillomas), and breast carcinomas (Fig. 24). The majority of breast carcinomas demonstrated high methylation of PRDMl 4, while only one out of 10 normal breast tissue samples, and a few benign tumors had low level methylation. Based on these data, PRDM 14 is a candidate biomarker for breast cancer diagnosis since it is methylated in 90% of invasive tumors and only 10% of normal breast tissue.

In addition, a MSP analysis of genomic DNA from a variety of pancreatic, prostate, lung, and breast cancer samples indicated that the PRDM14 gene is hypermethylated in a wide range of cancers (Table 13). Bisulfite treated DNA from the various cancer and normal tissues was amplified with: (a) a pair of PCR primers that effectively anneals only to methylated target sequences and causes the production of a detectable PCR product; and (b) and pair of primers that effectively only anneals to unmethylated target sequences and causes the production of a detectable PCR product.

Table 13.

Methylation of the PRDM14 gene in pancreatic, prostatic, lung,and breast cancer.

N, normal tissue from a healthy person (not a cancer patient).

N in CA, normal tissue adjacent to cancer tissue.

CA, cancer tissue. - Xenograft, cancer tissue grown in nude mice.

U, PCR product was detectable (on electrophoretic gels) only in PCR with unmethylated target - specific PCR primers.

WM (weakly methylated), PCR product was detectable (on electrophoetic gels) in PCR with both methylated and unmethylated target-specific PCR primers, but the methylated primer specific PCR was weak compared to the other sample.

The numbers in the M, WM, M, and Total columns are the numbers of different samples tested. Example 6. Analysis of gene expression by quantitative RT-PCR (qRT-PCR)

To further characterize the effect of methylation changes on gene expression, the expression of selected genes in cells purified from normal breast tissue, and in situ and invasive breast carcinomas was analyzed by RT-PCR (Figs. 25 A-D). Of the four genes analyzed both for methylation and gene expression, only one (Cxorfl2) had the differentially methylated sites localized in the predicted promoter area, while in the other three genes (PRDM14, H0XD4, and CDC42EP5) the differentially methylated Ascl and surrounding CpG sites were located in an intron or distal exon. Consistent with these findings, the relative expression of Cxorfl2 was positively correlated with methylation, while that of the other three genes was inversely correlated methylation. Thus, in all cases there was a strong correlation between differential methylation of the genes and their differential expression, but only methylation in the promoter area was associated with down-regulation of expression; in other regions it correlated with higher mRNA levels. These results are consistent with prior reports indicating that methylation in non- core (i.e., outside of the promoter) regions do not negatively affect transcription [Ushijima (2005) Nat. Rev. Cancer 5:223-231] and in some cases (e.g. H19/IGF2, an imprinted gene) DNA methylation in an intron leads to increased gene expression [Feinberg et al. (2004) Nat. Rev. Cancer 4:143-153; Bell et al. (2000) Nature 405, 482- 485]. The imprinting of IGF2 is dependent on CTCF binding to an enhancer-blocking element within the Hl 9 gene, the methylation of which inhibits CTCF binding and leads to loss of imprinting (LOI) [Feiber et al. (2004) supra; Bell et al. (2000) supra]. Interestingly, the differentially methylated regions identified in the PRDM 14 and CDC42EP5 genes (see above) appear to have a CTCF binding site [Bell et al. (2000) supra]. Thus, some of the genes identified herein are potentially subject to imprinting and the results presented above indicate possible loss of imprinting in a cell type and tumor stage specific manner.

In summary, a novel sequence-based method (Methylation Specific Digital Karyotyping; MSDK) for the analysis of the genome- wide methylation profiles is provided. MSDK analysis of three cell types (epithelial and myoepithelial cells and stromal fibroblasts) from normal breast tissue and in situ and invasive breast carcinomas revealed that distinct epigenetic changes occur in all three cell types during breast tumorigenesis. Alterations in stromal and myoepithelial cells thus likely play a role in the establishment of the abnormal tumor microenvironment and contribute to tumor progression. A number of embodiments of the invention have been described. Nevertheless, it will be understood that various modifications may be made without departing from the spirit and scope of the invention. Accordingly, other embodiments are within the scope of the following claims.

Example 7. Determination of the Global DNA Methylation of Stem Cells and

Their Differentiated Progeny

To determine the global methylation profile of putative normal mammary epithelial stem cells and their differentiated progeny, cells were purified from normal human breast tissue using known cell type specific cell surface markers (see Fig. 26A). Mammary epithelial stem cells were identified as lineage7CD24^"/low/CD44⁺ cells, while differentiated luminal epithelial cells were purified using anti-MUCl and anti-CD24 antibodies, and myoepithelial cells were isolated using anti-CD 10 antibodies. Hereafter, the putative normal mammary epithelial stem cells are referred to as CD44+ cells, the luminal epithelial cells as MUC 1+ or CD24+ cells, and myoepithelial cells as CD 10+ cells. The purity and differentiation status of the cells was confirmed by analyzing the expression of known differentiated (e.g., MUCl, MME) and mammary stem cell (e.g., IGFBP7, LRPl) markers by semi-quantitative RT-PCR (see Fig. 26B). SAGE (Serial Analysis of Gene Expression) libraries were also generated from each cell fraction to analyze their global expression profile. The SAGE data further confirmed the hypothesis that CD44+ cells represent stem cells while MUC 1+, CD24+, and CD 10+ cells represent a differentiated lineage of committed cells, since known luminal and myoepithelial lineage specific and stem markers were found mutually exclusively in the respective SAGE libraries. Example 8. Analysis of MSDK Data Obtained from Isolated Stem Cells and Their Differentiated Progeny

MSDK libraries were generated using genomic DNA isolated from CD44+, CD24+, MUC1+, and CD10+ cells purified as described above (see Figs. 26A and 26B). By comparing the actual number of MSDK tags obtained in each library to the expected or predicted number of MSDK tags, normal mammary epithelial stem cells (CD44+) were found to be hypomethylated compared to luminal epithelial (CD24+ or MUC 1+) and myoepithelial (CD10+) cells (see Table 14). Table 15 lists tags statistically significantly (p<0.05) differentially present in the four MSDK libraries.

Table 14. Chromosomal location and analysis of the frequency of MSDK tags in Stem and Differentiated Cells.

15

25

35

The column headings are as indicated for Table I₅ for the indicated purified cell populations, CDlO, CD24, CD44, and MUCl.

Table 15. List of tags statistically significantly (p<0.05) differentially present in the four Stem and Differentiated Cell MSDK libraries.

P value, the significance of the difference in the raw abundances of the relevant MSDK tag between the four libraries.

SEQ ID NO:, refers to the Sequence Identification Number assigned to each MSDK-tag nucleotide sequence

CDlO, CD24, CD44, MUCl, refer to the different cell populations used in the MSDK analysis.

Ascl position, refers to the bp position within the corresponding chromosome(s) where the Ascl site is located.

Chr, chromosome in which MSDK tag sequence is located.

UpGene, refers to nearest gene 5 ' to the Ascl site.

DnGene, refers to the nearest gene 3 ' to the Ascl site.

In addition, CDIO+and MUC1+ cells were also found to be hypomethylated compared to CD24+ cells. This latter observation raised the hypothesis (also suggested by SAGE data on these cells) that CD10+ and MUC1+ cells may represent a mix of terminally differentiated myoepithelial and luminal epithelial cells, respectively, and their lineage committed progenitors, while CD24+ cells are mostly terminally differentiated luminal epithelial cells. To identify loci specifically methylated in stem or differentiated cells of a specific lineage (luminal or myoepithelial), pair-wise as well as combined comparisons of the MSDK libraries were performed. Statistically significant (p<0.05) differences were found in each of these comparisons and led to the identification of tags that were specifically methylated in differentiated (luminal or myoepithelial) cells (see Figure 26C). Interestingly, many of the genes hypomethylated in CD44+ cells encode homeogenes, polycomb (chromo domain containing) proteins, or proteins involved in pathways known to be important for stem cell function. A detailed summary of these genes is shown in Table 16.

f

Table 16. Selected Differentially Methylated Genes in the CD44+ and CD24+ Libraries

CD24 and CD44, refer to the different cell populations (e.g., stem cell and differentiated cell populations) used in the MSDK analysis.

Chr, chromosome in which MSDK tag sequence is located. Gene, refers to nearest gene to the Ascl site. Position, refers to the location of the Ascl site within the associated gene, (i.e., Upstream (5') or inside (within the intronic or exonic portion of the gene).

Distance, refers to the distance of the Ascl site from the start site of transcription for the associated gene. Function, refers to the putative function associated with each gene located near the respective Ascl site.

Example 9. Confirmation of Stem and Differentiated Cell MSDK Results by

Bisulfite Sequencing Analysis

To confirm the MSDK results, sets of statistically significantly differentially methylated genes from each comparison were selected and their methylation status was analyzed by sequence analysis of bisulfite treated genomic DNA from the same sample that was used for MSDK. These genes included FNDCl and FOXCl (hypomethylated in CD44+ cells compared to all others), PACAP (hypomethylated in CD44+ and CD 10+ cells compared to others), SLC9A3R1 (hypomethylated in CD24+ MUC1+ and CD10+ cells compared to CD44+), DDNl (hypomethylated in CD44+ compared to CD 10+ cells), and DTXl and CDC42EP5 (hypomethylated in CD10+ compared to CD44+ cells). In all these cases, bisulfite sequencing analysis confirmed the MSDK results (see Fig. 27A).

Example 10. Determination of the Frequency and Consistency of Methylation Difference Between Stem and Differentiated Cells by qMSP

To determine how consistently the selected genes of Fig. 27 A are differentially methylated in stem and differentiated cells from multiple independent women, the quantitative methylation specific PCR (qMSP) assay (described above) was utilized to analyze methylation in a larger set of samples. qMSP confirmed MSDK and bisulfite sequencing data and demonstrated that cell lineage specific methylation is consistent among samples derived from women of different ages (18-58 years old) and reproductive history, although some variability in the degree of methylation was observed (see Fig. 27B).

Example 11. Analysis of Gene Expression of Selected Genes Differentiatially

Methylated in Stem and Differentiated Cells by qRT-PCR

To characterize the effect of methylation changes on gene expression, the expression of the selected genes was analyzed by quantitative RT-PCR in the same cells that were analyzed by qMSP in Example 10. Figure 28 shows the relative expression of the selected genes differentiatially methylated in CD44+, CD 10+, MUC 1+, and CD24+ cell subsets. Overall, an association between the methylation status and expression of the genes was observed. However, methylation did not have the same effect on expression of all the genes. The expression of FNDCl, DDN, LHXl, and HOXAlO was lower in methylated samples, while PACAP and CDC42EP5 were expressed at higher levels in hypermethylated cells. In the case of FOXCl and SOXl 3 in the CD44+, MUC 1+, and CD24+ samples, there was an inverse association between methylation and gene expression, but FOXCl was expressed in CD 10+ cells despite being methylated and SOXl 3 was not highly expressed in CD 10+ cells despite being hypomethylated. These variations could result if the CD 10+ cell fraction is a mix of myoepithelial progenitor and committed myoepithelial cells, and thus, has both progenitor and differentiated cell properties.

Example 12. Correlation of Methylation Status to Clinico-Pathologic Characteristics of Breast Carcinomas

To determine if the methylation of the most highly cell lineage specifically methylated genes would correlate with clinico-pathologic characteristics of breast carcinomas, the methylation of PACAP, FOXCl (both unmethylated in CD44+ cells compared to MUC1,CD24+ and CD10+ cells), and SLC9A3R1 (hypermethylated in CD44+ cells compared to all three other cell types) were analyzed in 149 sporadic invasive ductal carcinomas, 11 BRCAl⁺ tumors, 21 BRCA2⁺ tumors, and 14 phyllodes tumors. Based on this analysis, the methylation of PACAP and FOXCl were found to be statistically significantly associated with hormone receptor (estrogen receptor-ER, progesterone receptor-PR) and HER2 status of the tumors and with tumor subtypes. Basal-like tumors (defined as ER7PR7HER2^") and BRCAl tumors exhibited the same methylation profile as normal CD44+ stem cells, while ER⁺ and HER2⁺ tumors were more similar to differentiated cells. These results supported the hypothesis that either (a) different tumor subtypes have distinct cells of origin or (b) cancer stem cells in different tumors have different differentiation potential.

To evaluate these two hypotheses, qMSP analyses of putative cancer stem (Hn^" /CD24^"/l0W/CD44⁺/EPCR⁺) and differentiated cells (CD24+) cells were performed using genes that were highly cell type specifically methylated in normal breast tissue (see Fig. 29A). This analysis demonstrated that the DNA methylation profiles of tumor stem (CD44+) and CD24+ cells were the same as their corresponding normal counterparts, suggesting that regardless of the tumor subtype, cancer stem cells are likely to be more similar to each other and to normal stem cells than to more differentiated (CD24+) cells from the same tumor.

Example 13. Correlation of Methylation Status to Clinico-Pathologic

Characteristics of Breast Carcinomas

Based on the hypothesis that cancer stem cells are responsible for the metastatic spread and recurrence of tumors, the number of cancer stem cells would be expected to be higher in distant metastases compared to primary tumors. To test this hypothesis, the methylation status of four of the most highly cell type specifically methylated genes in primary tumors and matched distant metastases (collected from the same patient) was analyzed. Unexpectedly, the methylation of HOXAlO, FOXCl, and LHXl was higher in distant metastases compared to primary tumors, approaching or even exceeding levels detected in differentiated CD24+ cells, while no clear pattern was observed for PACAP (see Fig. 29B). This suggested that the number of CD24+ cells is increased in the distant metastasis, a finding reinforced by immunohistochemical analyses of these samples using stem and differentiated cell markers. Of the several plausible explanations of these results, the most likely is cell plasticity and different selection conditions in the primary tumor and distant metastases. Indeed, analysis of E-cadherin methylation and expression demonstrated that cell differentiation is a dynamic process and could occur during the metastatic progression. Thus, it is possible that the CD44+ cancer stem cells were the ones that metastasize, but they differentiate at the site of metastasis. Analysis of the genetic composition of CD24+ and CD44+ cells at the single cell level in primary tumors and matched metastases would be necessary to decipher this question.

In summary, the genome- wide DNA methylation profile of human putative mammary epithelial stem cells and differentiated luminal and myoepithelial cells was determined. Genes that were found to be methylated in a cell type specific manner demonstrated that cancer stem and differentiated cells are epigenetically distinct and are more similar to their corresponding normal counterparts than to each other, and the methylation status of selected genes classified breast tumors into cell subtypes.

Claims

WHAT IS CLAIMED IS:

1. A method of making a methylation specific digital karyotyping (MSDK) library, the method comprising:

5 providing all or part of the genomic DNA of a test cell; exposing the DNA to a methylation-sensitive mapping restriction enzyme (MMRE) to generate a plurality of first fragments; conjugating to one terminus or to both termini of each of the first fragments a binding moiety, the binding moiety comprising a first member of an affinity o pair, the conjugating resulting in a plurality of second fragments; exposing the plurality of second fragments to a fragmenting restriction enzyme (FRE) to generate a plurality of third fragments, each third fragment comprising at one terminus the first member of the affinity pair and at the other terminus the 5' cut sequence of the FRE or the 3' cut sequence of the FRE; 5 contacting the plurality of third fragments with an insoluble substrate having bound thereto a plurality of second members of the affinity pair, said contacting resulting in a plurality of bound third fragments, each bound third fragment being a third fragment bound via the first and second members of the affinity pair to the insoluble substrate; 0 conjugating to free termini of the bound third fragments a releasing moiety, the releasing moiety comprising a releasing restriction enzyme (RRE) recognition sequence and, 3' of the recognition sequence of the RRE, either the 5' cut sequence of the FRE or the 3' cut sequence of the FRE, the conjugating resulting in a plurality of bound fourth fragments, each bound fourth fragment (i) comprising at one 5 terminus the recognition sequence of the RRE and (ii) being bound via the first member of the affinity pair at the other terminus and the second member of the affinity pair to the insoluble substrate; and exposing the bound fourth fragments to the RRE, the exposing resulting in the release from the insoluble substrate of a MSDK library, the library comprising a plurality of fifth fragments, each fifth fragment comprising the releasing moiety and a MSDK tag, the tag consisting of a plurality of base pairs of the genomic DNA.

2. The method of claim 1 , wherein the MMRE is Ascl.

3. The method of claim 1 , wherein the FRE is NIaIII.

4. The method of claim 1, wherein the RRE is Mmel.

5. The method of claim 1, wherein the binding moiety further comprises a 5' or 3' cut sequence of the MMRE.

6. The method of claim 1 , wherein the binding moiety further comprises, between the 5 ' or 3' recognition sequence of the MMRE and the first member of an affinity pair, a linker nucleic acid sequence comprising a plurality of base pairs.

7. The method of claim 1, wherein the releasing moiety further comprises, 5' of the RRE recognition sequence, an extender nucleic acid sequence comprising a plurality of base pairs.

8. A method of analyzing a MSDK library, the method comprising; providing a MSDK library made by the method of claim 1 ; identifying the nucleotide sequences of one tag, a plurality of tags, or all of the tags.

9. The method of claim 8, wherein identifying the nucleotide sequences of a plurality of tags comprises: making a plurality of ditags, each ditag comprising two fifth fragments ligated together; forming a concatamer comprising a plurality of ditags or ditag fragments, wherein each ditag fragment comprises two MSDK tags; determining the nucleotide sequence of the concatamer; and deducing, from the nucleotide sequence of the concatamer, the nucleotide sequences of one or more of the MSDK tags that the concatamer comprises.

10. The method of claim 9, wherein the ditag fragments are made by exposing ditags to the FRE.

11. The method of claim 9, further comprising, after making a plurality of ditags and prior to forming the concatamers, increasing the number of ditags by PCR.

12. The method of claim 8, further comprising determining the relative frequency of some or all of the tags.

13. A method of analyzing a MSDK library, the method comprising: providing a MSDK library made by the method of claim 1 ; and identifying a chromosomal site corresponding to the sequence of a tag selected from the library.

14. The method of claim 9, further comprising determining a chromosomal location, in the genome of the test cell, of an unmethylated full recognition sequence of the MMRE closest to the identified chromosomal site.

15. The method of claim 13, wherein the identification of the chromosomal site and the determination of the chromosomal location is performed by a process comprising comparing the nucleotide sequence of the selected tag to a virtual tag library generated using the nucleotide sequence of the genome or the part of a genome, the nucleotide sequence of the full recognition sequence of the MMRE, the nucleotide sequence of the full recognition sequence of the FRE, and the number of nucleotides separating the full recognition sequence of the RRE from the RRE cutting site.

16. A method of determining the chromosomal location of a plurality of unmethylated recognition sequences of the MMRE, the method comprising repeating the method of claim 14 with a plurality of tags obtained from the library.

17. The method of claim 1, wherein the test cell is a vertebrate cell.

18. The method of claim 1, wherein the test cell is a mammalian test cell.

19. The method of claim 18, wherein the mammalian test cell is a human test cell.

20. The method of claim 18, wherein the test cell is a normal cell.

21. The method of claim 18, wherein the test cell is a cancer cell.

22. The method of claim 21, wherein the cancer cell is a breast cancer cell.

23. The method of claim 1 , wherein the first member of the affinity pair is biotin or iminobiotin.

24. The method of claim 1, wherein the first member of the affinity pair is an antigen, a haptenic determinant, a single-stranded nucleotide sequence, a hormone, a ligand for adhesion receptor, a receptor for an adhesion ligand, a ligand for a lectin, a lectin, a molecule containing all or part of an immunoglobulin Fc region, bacterial protein A, or bacterial protein G.

25. The method of claim 1, wherein the insoluble substrate comprises magnetic beads.

26. A method of classifying a biological cell, the method comprising: (a) performing the method of claim 12, thereby obtaining a test MSDK profile for the test cell;

(b) comparing the test MSDK profile to separate control MSDK expression profiles for one or more control cell types; (c) selecting a control MSDK profile that most closely resembles the test

MSKD profile; and

(d) assigning to the test cell a cell type that matches the cell type of the control MSDK profile selected in step (c).

27. The method of claim 26, wherein the test and control cells are vertebrate cells.

28. The method of claim 27, wherein the test and control cells are mammalian cells.

29. The method of claim 28, wherein the test and control cells are human cells.

30. The method of claim 28, wherein the control cell types comprise a control normal cell and a control cancer cell of the same tissue as the normal cell.

31. The method of claim 30, wherein the control normal cell and the control cancer cell are breast cells.

32. The method of claim 30, wherein the control normal cell and the control cancer cell are of a tissue selected from the group consisting of colon, lung, prostate, and pancreas.

33. The method of claim 30, wherein the test cell is a breast cell.

34. The method of claim 30, wherein the test cell is of a tissue selected from the group consisting of colon, lung, prostate, and pancreas.

35. The method of claim 26, wherein the control cell types comprise cells of different categories of a cancer of a single tissue.

36. The method of claim 35, wherein the different categories of a cancer of a single tissue comprise a breast ductal carcinoma in situ (DCIS) cell and an invasive breast cancer cell.

37. The method of claim 35 , wherein the different categories of a cancer of a single tissue comprise two or more of: a high grade DCIS cell, an intermediate grade DCIS cell; and an low grade DCIS cell.

38. The method of claim 28, wherein the control cell types comprise two or more of: a lung cancer cell; a breast cancer cell; a colon cancer cell; a prostate cancer cell; and a pancreatic cancer cell.

39. The method of claim 26, wherein the control cell types comprise an epithelial cell obtained from non-cancerous tissue and a myoepithelial cell obtained from non-cancerous tissue.

40. A method of diagnosis, the method comprising: (a) providing a test breast epithelial cell; (b) determining the degree of methylation of one or more C residues in a gene in the test cell, wherein the gene is selected from those identified by the MSDK tags listed in Table 5, wherein the one or more C residues are C residues in CpG sequences; and

(c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control epithelial cell obtained from non-cancerous breast tissue, wherein an altered degree of methylation of the one or more C residues in the test epithelial cell compared to the control epithelial cell is an indication that the test epithelial cell is a cancer cell.

41. The method of claim 40, wherein the altered degree of methylation is a lower degree of methylation.

42. The method of claim 40, wherein the altered degree of methylation is a higher degree of methylation.

43. The method of claim 40, wherein the altered degree of methylation is in the promoter region of the gene.

44. The method of claim 40, wherein the altered degree of methylation is in an exon of the gene or an intron of the gene.

45. The method of claim 40, wherein the gene is selected from the group consisting of PRDM14 and ZCCHC14.

46. A method of diagnosis, the method comprising: (a) providing a test colon epithelial cell;

(b) determining the degree of methylation of one or more C residues in a gene in the test cell, wherein the gene is selected from those identified by the MSDK tags listed in Table 2, wherein the one or more C residues are C residues in CpG sequences; and (c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control epithelial cell obtained from non-cancerous colon tissue, wherein an altered degree of methylation of the one or more C residues in the test epithelial cell compared to the control epithelial cell is an indication that the test epithelial cell is a cancer cell.

47. The method of claim 46, wherein the altered degree of methylation is a lower degree of methylation.

48. The method of claim 46, wherein the altered degree of methylation is a higher degree of methylation.

49. The method of claim 46, wherein the altered degree of methylation is in the promoter region of the gene.

50. The method of claim 46, wherein the altered degree of methylation is in an exon of the gene or an intron of the gene.

51. The method of claim 46, wherein the gene is selected from the group consisting of LHX3, TCF7L1, and LMX-IA.

52. A method of diagnosis, the method comprising:

(a) providing a test myoepithelial cell obtained from a test breast tissue;

(b) determining the degree of methylation of one or more C residues in a gene in the test cell, wherein the gene is selected from those identified by the MSDK tags listed in Table 10, wherein the one or more C residues are C residues in CpG sequences; and

(c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control myoepithelial cell obtained from non-cancerous breast tissue, wherein an altered degree of methylation of the one or more C residues in the test myoepithelial cell compared to the control myoepithelial cell is an indication that the test breast tissue is cancerous tissue.

53. The method of claim 52, wherein the altered degree of methylation is a lower degree of methylation.

54. The method of claim 52, wherein the altered degree of methylation is a higher degree of methylation.

55. The method of claim 52, wherein the altered degree of methylation is in the promoter region of the gene.

56. The method of claim 52, wherein the altered degree of methylation is in an exon of the gene or an intron of the gene.

57. The method of claim 52, wherein the gene is selected from the group consisting of H0XD4, SLC9A3R1, and CDC42EP5.

58. A method of diagnosis, the method comprising:

(a) providing a test fibroblast obtained from a test breast tissue; (b) determining the degree of methylation of one or more C residues in a gene in the test cell, wherein the gene is selected from those identified by the MSDK tags listed in Tables 7 and 8, wherein the one or more C residues are C residues in CpG sequences; and

(c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control fibroblast obtained from non-cancerous breast tissue, wherein an altered degree of methylation of the one or more C residues in the test fibroblast compared to the control fibroblast is an indication that the test breast tissue is cancerous tissue.

59. The method of claim 58, wherein the altered degree of methylation is a lower degree of methylation.

60. The method of claim 58, wherein the altered degree of methylation is a higher degree of methylation.

61. The method of claim 58 wherein the altered degree of methylation is in the promoter region of the gene.

62. The method of claim 58, wherein the altered degree of methylation is in an 5 exon of the gene or an intron of the gene.

63. The method of claim 58 wherein the gene is Cxorfl2.

64. A method of determining the likelihood of a cell being an epithelial cell or o a myoepithelial cell, the method comprising:

(a) providing a test cell;

(b) determining the degree of methylation of one or more C residues in a gene in the test cell, wherein the gene is selected from those identified by the MSDK tags listed in Table 12, wherein the one or more C residues are C residues in CpG sequences; 5 and

(c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control myoepithelial cell and to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control epithelial cell, wherein the test cell 0 is: (i) more likely to be a myoepithelial cell if the degree of methylation in the test sample more closely resembles the degree of methylation in the control myoepithelial cell; or (ii) more likely to be an epithelial cell if the degree of methylation in the test sample more closely resembles the degree of methylation in the control epithelial cell.

5 65. The method of claim 64, wherein the C residues are in the promoter region of the gene.

66. The method of claim 64, wherein the C residues are in an exon of the gene or an intron of the gene. 0

67. The method of claim 64, wherein the gene is selected from the group consisting of LOC389333 and CDC42EP5.

68. A method of diagnosis, the method comprising: (a) providing a test cell from a test tissue;

(b) determining the degree of methylation of one or more C residues in a PRDM14 gene in the test cell, wherein the one or more C residues are C residues in CpG sequences; and

(c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in the PRDM14 gene in a control cell obtained from non-cancerous tissue of the same tissue as the test cell, wherein an altered degree of methylation of the one or more C residues in the test cell compared to the control cell is an indication that the test cell is a cancer cell.

69. The method of claim 68, wherein the altered degree of methylation is a lower degree of methylation.

70. The method of claim 68, wherein the altered degree of methylation is a higher degree of methylation.

71. The method of claim 68, wherein the altered degree of methylation is in the promoter region of the gene.

72. The method of claim 68, wherein the altered degree of methylation is in an exon of the gene or an intron of the gene.

73. The method of claim 68, wherein the test and control cell are breast cells.

74. The method of claim 68, wherein the test and control cells are of a tissue selected from the group consisting of colon, lung, prostate, and pancreas.

75. A method of diagnosis comprising:

(a) providing a test sample of breast tissue comprising a test epithelial cell;

(b) determining the level of expression in the test epithelial cell of a gene selected from those listed in Table 5, wherein the gene is one that is expressed in a breast cancer epithelial cell at a substantially altered level compared to a compared to a normal breast epithelial cell; and

(c) classifying the test cell as: (i) a normal breast epithelial cell if the level of expression of the gene in the test cell is not substantially altered compared to a control level of expression for a normal breast epithelial cell; or (ii) a breast cancer epithelial cell if the level of expression of the gene in the test cell is substantially altered compared to a control level of expression for a normal breast epithelial cell.

76. The method of claim 75, wherein the gene is selected from the group consisting of PRDM14 and ZCCHC14.

77. The method of claim 75, wherein the alteration in the level of expression is an increase in the level of expression.

78. The method of claim 75, wherein the alteration in the level of expression is a decrease in the level of expression.

79. A method of diagnosis comprising:

(a) providing a test sample of colon tissue comprising a test epithelial cell;

(b) determining the level of expression in the test epithelial cell of a gene selected from those listed in Table 2, wherein the gene is one that is expressed in a colon cancer epithelial cell at a substantially altered level compared to a compared to a normal colon epithelial cell; and

(c) classifying the test cell as: (i) a normal colon epithelial cell if the level of expression of the gene in the test cell is not substantially altered compared to a control level of expression for a normal colon epithelial cell; or (ii) a colon cancer epithelial cell if the level of expression of the gene in the test cell is substantially altered compared to a control level of expression for a normal colon epithelial cell.

80. The method of claim 79, wherein the gene is selected from the group consisting of LHX3, TCF7L1, and LMX-IA.

81. The method of claim 79, wherein the alteration in the level of expression is an increase in the level of expression.

82. The method of claim 79, wherein the alteration in the level of expression is a decrease in the level of expression.

83. A method of diagnosis comprising:

(a) providing a test sample of breast tissue comprising a test stromal cell; (b) determining the level of expression in the stromal cell of a gene selected from those listed in Tables 7, 8, and 10, wherein the gene is one that is expressed in a cell of the same type as the test stromal cell at a substantially altered level when present in breast cancer tissue than when present in normal breast tissue; and

(c) classifying the test sample as: (i) normal breast tissue if the level of expression of the gene in the test stromal cell is not substantially altered compared to a control level of expression for a control cell of the same type as the test stromal cell in normal breast tissue; or (ii) breast cancer tissue if the level of expression of the gene in the test stromal cell is substantially altered compared to a control level of expression for a control cell of the same type as the test stromal cell in normal breast tissue.

84. The method of claim 83, wherein the test and control stromal cells are myoepithelial cells and the genes are selected from those listed in Table 10.

85. The method of claim 83, wherein the gene is selected from the group consisting of HOXD4, SLC9A3R1, and CDC32EP5.

86. The method of claim 83, wherein the test and control stromal cells are fibroblasts and the genes are selected from those listed in Tables 7 and 8.

87. The method of claim 83, wherein the gene is Cxorf 12.

88. The method of claim 83, wherein the alteration in the level of expression is an increase in the level of expression.

89. The method of claim 83, wherein the alteration in the level of expression is a decrease in the level of expression.

90. A method of determining the likelihood of a cell being an epithelial cell or a myoepithelial cell, the method comprising:

(a) providing a test cell; (b) determining the level of expression in the test sample of a gene selected from the group consisting of those identified by the MSDK tags listed in Table 12;

(c) determining whether the level of expression of the selected gene in the test sample more closely resembles the level of expression of the selected gene in (i) a control myoepithelial cell or (ii) a control epithelial cell; and

(d) classifying the test cell as: (i) likely to be a myoepithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control myoepithelial cell; or (ii) likely to be an epithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control epithelial cell.

91. The method of claim 90, wherein the gene is selected from the group consisting of LOC389333 and CDC42EP5.

92. A method of diagnosis comprising:

(a) providing a test cell; (b) determining the level of expression in the test cell of a PRDM 14 gene; and

(c) classifying the test cell as: (i) a normal cell if the level of expression of the gene in the test cell is not substantially altered compared to a control level of expression for a control normal cell of the same tissue as the test cell; or (ii) a cancer cell if the level of expression of the gene in the test cell is substantially altered compared to a control level of expression for a control normal cell of the same tissue as the test cell.

93. The method of claim 92, wherein the alteration in the level of expression is an increase in the level of expression.

94. The method of claim 92, wherein the alteration in the level of expression is a decrease in the level of expression.

95. The method of claim 92, wherein the test and control cells are breast cells.

96. The method of claim 92, wherein the test and control cells are of a tissue selected from the group consisting of colon, lung, prostate, and pancreas.

97. A single stranded nucleic acid probe comprising:

(a) the nucleotide sequence of a tag selected from those listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16; or

(b) the complement of the nucleotide sequence.

98. An array comprising a substrate having at least 10 addresses, wherein each address has disposed thereon a capture probe comprising:

(a) a nucleic acid sequence consisting of a tag nucleotide sequence selected from those listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16; or

(b) the complement of the nucleic acid sequence.

99. A kit comprising at least 10 probes, each probe comprising: (a) a nucleic acid sequence comprising a tag nucleotide sequence selected from those listed in Tables 2, 5, 7, 8, 10, 12, 15 and 16; or

Qo) the complement of the nucleic acid sequence.

100. A kit comprising at least 10 antibodies each of which is specific for a different protein encoded by a gene identified by a tag selected from the group consisting of the tags listed in Tables 2, 5, 7, 8, 10, 12, 15, and 16.

101. A method of determining the likelihood of a cell being a stem cell, an differentiated luminal epithelial cell or a myoepithelial cell, the method comprising:

(a) providing a test cell;

(b) determining the degree of methylation of one or more C residues in a gene in the test cell, wherein the gene is selected from those identified by the MSDK tags listed in Table 15 or 16, wherein the one or more C residues are C residues in CpG sequences; and

(c) comparing the degree of methylation of the one or more residues to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control stem cell, to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control stem cell, and to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control differentiated luminal epithelial cell, and to the degree of methylation of corresponding one or more C residues in a corresponding gene in a control myoepithelial cell, wherein the test cell is: (i) more likely to be a stem cell if the degree of methylation in the test cell more closely resembles the degree of methylation in the control stem cell; (ii) more likely to be a differentiated luminal epithelial cell if the degree of methylation in the test cell more closely resembles the degree of methylation in the control differentiated luminal epithelial cell; or (iii) more likely to be an myoepithelial cell if the degree of methylation in the test cell more closely resembles the degree of methylation in the control myoepithelial cell

102. The method of claim 101, wherein the C residues are in the promoter region of the gene.

103. The method of claim 101, wherein the C residues are in an exon of the gene or an intron of the gene.

104. The method of claim 101, wherein the gene is selected from the group consisting of SOX13, SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and HOXAlO.

105. A method of determining the likelihood of a cell being a stem cell, a differentiated luminal epithelial cell, or a myoepithelial cell, the method comprising:

(a) providing a test cell;

(b) determining the level of expression in the test sample of a gene selected from the group consisting of those identified by the MSDK tags listed in Table 15 or 16;

(c) determining whether the level of expression of the selected gene in the test sample more closely resembles the level of expression of the selected gene in (i) a control stem cell, (ii) a control differentiated luminal epithelial cell, or (ii) a control myoepithelial cell; and

(d) classifying the test cell as: (i) likely to be a stem cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control stem cell; (ii) likely to be a differentiated luminal epithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control epithelial cell; or (iii) likely to be an myoepithelial cell if the level of expression of the gene in the test cell more closely resembles the level of expression of the gene in a control myoepithelial cell.

106. The method of claim 105, wherein the control cells are breast cells.

107. The method of claim 106, wherein the breast cells are breast cancer cells.

108. The method of any of claims 105-107, wherein the gene is selected from the group consisting of SOX13, SLC9A3R1, FNDCl, FOXCl, PACAP, DDN, CDC42EP5, LHXl, and HOXAlO.